Add files using upload-large-folder tool

l2-13b-ga/checkpoint-1300/config.json

@@ -0,0 +1,30 @@
+{
+  "_name_or_path": "meta-llama/Llama-2-13b-hf",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 5120,
+  "initializer_range": 0.02,
+  "intermediate_size": 13824,
+  "max_position_embeddings": 4096,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 40,
+  "num_hidden_layers": 40,
+  "num_key_value_heads": 40,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.46.3",
+  "use_cache": true,
+  "vocab_size": 35483
+}

l2-13b-ga/checkpoint-1300/generation_config.json

@@ -0,0 +1,10 @@
+{
+  "bos_token_id": 1,
+  "do_sample": true,
+  "eos_token_id": 2,
+  "max_length": 4096,
+  "pad_token_id": 0,
+  "temperature": 0.6,
+  "top_p": 0.9,
+  "transformers_version": "4.46.3"
+}

l2-13b-ga/checkpoint-1300/latest

@@ -0,0 +1 @@
+global_step1300

l2-13b-ga/checkpoint-1300/model.safetensors.index.json

@@ -0,0 +1,370 @@
+{
+  "metadata": {
+    "total_size": 26103060480
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00006-of-00006.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.15.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.30.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.31.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.input_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.down_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.gate_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.up_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.post_attention_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.39.input_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.down_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.gate_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.up_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.post_attention_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.k_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.o_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.q_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.v_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.norm.weight": "model-00006-of-00006.safetensors"
+  }
+}

l2-13b-ga/checkpoint-1300/special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

l2-13b-ga/checkpoint-1300/tokenizer_config.json

@@ -0,0 +1,42 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": true,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": true,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

l2-13b-ga/checkpoint-1300/trainer_state.json

@@ -0,0 +1,950 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.2142190053700677,
+  "eval_steps": 500,
+  "global_step": 1300,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0009339248190520663,
+      "grad_norm": 6.638877692627699,
+      "learning_rate": 9.345794392523364e-07,
+      "loss": 9.2917,
+      "step": 1
+    },
+    {
+      "epoch": 0.009339248190520663,
+      "grad_norm": 1.1560921335705272,
+      "learning_rate": 9.345794392523365e-06,
+      "loss": 9.0876,
+      "step": 10
+    },
+    {
+      "epoch": 0.018678496381041326,
+      "grad_norm": 0.8415132296956432,
+      "learning_rate": 1.869158878504673e-05,
+      "loss": 8.2164,
+      "step": 20
+    },
+    {
+      "epoch": 0.02801774457156199,
+      "grad_norm": 0.45381630992958155,
+      "learning_rate": 2.8037383177570094e-05,
+      "loss": 7.5184,
+      "step": 30
+    },
+    {
+      "epoch": 0.03735699276208265,
+      "grad_norm": 0.8400636107958425,
+      "learning_rate": 3.738317757009346e-05,
+      "loss": 6.6507,
+      "step": 40
+    },
+    {
+      "epoch": 0.046696240952603316,
+      "grad_norm": 0.557696240829066,
+      "learning_rate": 4.672897196261683e-05,
+      "loss": 5.8909,
+      "step": 50
+    },
+    {
+      "epoch": 0.05603548914312398,
+      "grad_norm": 0.3971996057467842,
+      "learning_rate": 5.607476635514019e-05,
+      "loss": 5.4127,
+      "step": 60
+    },
+    {
+      "epoch": 0.06537473733364464,
+      "grad_norm": 0.2932710540265688,
+      "learning_rate": 6.542056074766355e-05,
+      "loss": 5.0106,
+      "step": 70
+    },
+    {
+      "epoch": 0.0747139855241653,
+      "grad_norm": 0.3682690443551033,
+      "learning_rate": 7.476635514018692e-05,
+      "loss": 4.6042,
+      "step": 80
+    },
+    {
+      "epoch": 0.08405323371468597,
+      "grad_norm": 0.3132971920011515,
+      "learning_rate": 8.411214953271028e-05,
+      "loss": 4.2031,
+      "step": 90
+    },
+    {
+      "epoch": 0.09339248190520663,
+      "grad_norm": 0.6731868159213446,
+      "learning_rate": 9.345794392523365e-05,
+      "loss": 3.9423,
+      "step": 100
+    },
+    {
+      "epoch": 0.1027317300957273,
+      "grad_norm": 0.27848867836763197,
+      "learning_rate": 0.000102803738317757,
+      "loss": 3.7157,
+      "step": 110
+    },
+    {
+      "epoch": 0.11207097828624796,
+      "grad_norm": 0.24642109032991807,
+      "learning_rate": 0.00011214953271028037,
+      "loss": 3.4516,
+      "step": 120
+    },
+    {
+      "epoch": 0.12141022647676862,
+      "grad_norm": 0.25717384664029797,
+      "learning_rate": 0.00012149532710280373,
+      "loss": 3.2167,
+      "step": 130
+    },
+    {
+      "epoch": 0.13074947466728928,
+      "grad_norm": 0.20912922668565637,
+      "learning_rate": 0.0001308411214953271,
+      "loss": 3.0237,
+      "step": 140
+    },
+    {
+      "epoch": 0.14008872285780993,
+      "grad_norm": 0.15805888388706113,
+      "learning_rate": 0.00014018691588785047,
+      "loss": 2.8529,
+      "step": 150
+    },
+    {
+      "epoch": 0.1494279710483306,
+      "grad_norm": 0.23370349497479534,
+      "learning_rate": 0.00014953271028037384,
+      "loss": 2.7078,
+      "step": 160
+    },
+    {
+      "epoch": 0.15876721923885126,
+      "grad_norm": 0.1802138633012483,
+      "learning_rate": 0.0001588785046728972,
+      "loss": 2.6115,
+      "step": 170
+    },
+    {
+      "epoch": 0.16810646742937194,
+      "grad_norm": 0.13354347610039718,
+      "learning_rate": 0.00016822429906542056,
+      "loss": 2.5309,
+      "step": 180
+    },
+    {
+      "epoch": 0.17744571561989259,
+      "grad_norm": 0.09414865188086892,
+      "learning_rate": 0.00017757009345794393,
+      "loss": 2.4452,
+      "step": 190
+    },
+    {
+      "epoch": 0.18678496381041326,
+      "grad_norm": 0.08333601554768896,
+      "learning_rate": 0.0001869158878504673,
+      "loss": 2.3832,
+      "step": 200
+    },
+    {
+      "epoch": 0.1961242120009339,
+      "grad_norm": 0.15926414699806835,
+      "learning_rate": 0.00019626168224299065,
+      "loss": 2.3492,
+      "step": 210
+    },
+    {
+      "epoch": 0.2054634601914546,
+      "grad_norm": 0.09492820761057012,
+      "learning_rate": 0.0001999989254250208,
+      "loss": 2.323,
+      "step": 220
+    },
+    {
+      "epoch": 0.21480270838197524,
+      "grad_norm": 0.0801349259356147,
+      "learning_rate": 0.00019999235866155886,
+      "loss": 2.2731,
+      "step": 230
+    },
+    {
+      "epoch": 0.22414195657249592,
+      "grad_norm": 0.12210960524693895,
+      "learning_rate": 0.00019997982251228469,
+      "loss": 2.2433,
+      "step": 240
+    },
+    {
+      "epoch": 0.23348120476301656,
+      "grad_norm": 3.14289498732125,
+      "learning_rate": 0.00019996131772558666,
+      "loss": 3.2769,
+      "step": 250
+    },
+    {
+      "epoch": 0.24282045295353724,
+      "grad_norm": 1.632940983166179,
+      "learning_rate": 0.00019993684540617132,
+      "loss": 4.9343,
+      "step": 260
+    },
+    {
+      "epoch": 0.2521597011440579,
+      "grad_norm": 3.4831252230225416,
+      "learning_rate": 0.00019990640701499736,
+      "loss": 4.2768,
+      "step": 270
+    },
+    {
+      "epoch": 0.26149894933457857,
+      "grad_norm": 1.6069045920523788,
+      "learning_rate": 0.00019987000436918874,
+      "loss": 5.9581,
+      "step": 280
+    },
+    {
+      "epoch": 0.27083819752509924,
+      "grad_norm": 0.2220907936615993,
+      "learning_rate": 0.00019982763964192585,
+      "loss": 3.8228,
+      "step": 290
+    },
+    {
+      "epoch": 0.28017744571561987,
+      "grad_norm": 0.24737284913291765,
+      "learning_rate": 0.00019977931536231596,
+      "loss": 3.1413,
+      "step": 300
+    },
+    {
+      "epoch": 0.28951669390614054,
+      "grad_norm": 4.010404518241152,
+      "learning_rate": 0.00019972503441524224,
+      "loss": 2.8432,
+      "step": 310
+    },
+    {
+      "epoch": 0.2988559420966612,
+      "grad_norm": 0.1515583580811596,
+      "learning_rate": 0.00019966480004119142,
+      "loss": 2.7859,
+      "step": 320
+    },
+    {
+      "epoch": 0.3081951902871819,
+      "grad_norm": 0.11259395750650594,
+      "learning_rate": 0.00019959861583606045,
+      "loss": 2.5821,
+      "step": 330
+    },
+    {
+      "epoch": 0.3175344384777025,
+      "grad_norm": 0.22514797814956813,
+      "learning_rate": 0.00019952648575094183,
+      "loss": 2.4517,
+      "step": 340
+    },
+    {
+      "epoch": 0.3268736866682232,
+      "grad_norm": 0.08040136172033542,
+      "learning_rate": 0.00019944841409188767,
+      "loss": 2.3794,
+      "step": 350
+    },
+    {
+      "epoch": 0.3362129348587439,
+      "grad_norm": 0.054758073593565354,
+      "learning_rate": 0.00019936440551965263,
+      "loss": 2.3232,
+      "step": 360
+    },
+    {
+      "epoch": 0.34555218304926455,
+      "grad_norm": 0.06742998909645591,
+      "learning_rate": 0.00019927446504941577,
+      "loss": 2.2776,
+      "step": 370
+    },
+    {
+      "epoch": 0.35489143123978517,
+      "grad_norm": 0.048780907584876736,
+      "learning_rate": 0.00019917859805048096,
+      "loss": 2.2376,
+      "step": 380
+    },
+    {
+      "epoch": 0.36423067943030585,
+      "grad_norm": 0.0475325963052214,
+      "learning_rate": 0.00019907681024595663,
+      "loss": 2.2191,
+      "step": 390
+    },
+    {
+      "epoch": 0.3735699276208265,
+      "grad_norm": 0.054089563211590065,
+      "learning_rate": 0.00019896910771241387,
+      "loss": 2.1961,
+      "step": 400
+    },
+    {
+      "epoch": 0.3829091758113472,
+      "grad_norm": 0.21798406131864823,
+      "learning_rate": 0.00019885549687952372,
+      "loss": 2.2078,
+      "step": 410
+    },
+    {
+      "epoch": 0.3922484240018678,
+      "grad_norm": 0.8673185709111124,
+      "learning_rate": 0.00019873598452967338,
+      "loss": 2.3731,
+      "step": 420
+    },
+    {
+      "epoch": 0.4015876721923885,
+      "grad_norm": 0.22424350669971718,
+      "learning_rate": 0.0001986105777975613,
+      "loss": 2.6195,
+      "step": 430
+    },
+    {
+      "epoch": 0.4109269203829092,
+      "grad_norm": 0.307418135168262,
+      "learning_rate": 0.00019847928416977126,
+      "loss": 2.3624,
+      "step": 440
+    },
+    {
+      "epoch": 0.42026616857342985,
+      "grad_norm": 0.07944722668080402,
+      "learning_rate": 0.00019834211148432536,
+      "loss": 2.2799,
+      "step": 450
+    },
+    {
+      "epoch": 0.4296054167639505,
+      "grad_norm": 0.18146933758664588,
+      "learning_rate": 0.00019819906793021614,
+      "loss": 2.2177,
+      "step": 460
+    },
+    {
+      "epoch": 0.43894466495447115,
+      "grad_norm": 0.07035825837333018,
+      "learning_rate": 0.0001980501620469178,
+      "loss": 2.1767,
+      "step": 470
+    },
+    {
+      "epoch": 0.44828391314499183,
+      "grad_norm": 0.04596186944454228,
+      "learning_rate": 0.0001978954027238763,
+      "loss": 2.1598,
+      "step": 480
+    },
+    {
+      "epoch": 0.4576231613355125,
+      "grad_norm": 0.041342347745088055,
+      "learning_rate": 0.0001977347991999786,
+      "loss": 2.131,
+      "step": 490
+    },
+    {
+      "epoch": 0.46696240952603313,
+      "grad_norm": 0.04172063219841485,
+      "learning_rate": 0.00019756836106300137,
+      "loss": 2.1231,
+      "step": 500
+    },
+    {
+      "epoch": 0.4763016577165538,
+      "grad_norm": 0.03373646457711144,
+      "learning_rate": 0.00019739609824903843,
+      "loss": 2.1146,
+      "step": 510
+    },
+    {
+      "epoch": 0.4856409059070745,
+      "grad_norm": 0.03736871030676605,
+      "learning_rate": 0.00019721802104190748,
+      "loss": 2.1003,
+      "step": 520
+    },
+    {
+      "epoch": 0.49498015409759516,
+      "grad_norm": 0.033931028038211034,
+      "learning_rate": 0.00019703414007253645,
+      "loss": 2.0983,
+      "step": 530
+    },
+    {
+      "epoch": 0.5043194022881158,
+      "grad_norm": 0.03790055446070549,
+      "learning_rate": 0.00019684446631832868,
+      "loss": 2.092,
+      "step": 540
+    },
+    {
+      "epoch": 0.5136586504786365,
+      "grad_norm": 0.030956192803893078,
+      "learning_rate": 0.00019664901110250758,
+      "loss": 2.0807,
+      "step": 550
+    },
+    {
+      "epoch": 0.5229978986691571,
+      "grad_norm": 0.03542530209935129,
+      "learning_rate": 0.00019644778609344068,
+      "loss": 2.0773,
+      "step": 560
+    },
+    {
+      "epoch": 0.5323371468596778,
+      "grad_norm": 0.040947757568902336,
+      "learning_rate": 0.00019624080330394306,
+      "loss": 2.0649,
+      "step": 570
+    },
+    {
+      "epoch": 0.5416763950501985,
+      "grad_norm": 0.034273415973688146,
+      "learning_rate": 0.00019602807509056018,
+      "loss": 2.0479,
+      "step": 580
+    },
+    {
+      "epoch": 0.5510156432407192,
+      "grad_norm": 0.031427481498873144,
+      "learning_rate": 0.00019580961415283028,
+      "loss": 2.0563,
+      "step": 590
+    },
+    {
+      "epoch": 0.5603548914312397,
+      "grad_norm": 0.03141549752041532,
+      "learning_rate": 0.00019558543353252611,
+      "loss": 2.0503,
+      "step": 600
+    },
+    {
+      "epoch": 0.5696941396217604,
+      "grad_norm": 0.033012392726428204,
+      "learning_rate": 0.00019535554661287652,
+      "loss": 2.0389,
+      "step": 610
+    },
+    {
+      "epoch": 0.5790333878122811,
+      "grad_norm": 0.02913261992661444,
+      "learning_rate": 0.0001951199671177673,
+      "loss": 2.036,
+      "step": 620
+    },
+    {
+      "epoch": 0.5883726360028018,
+      "grad_norm": 0.030543903708435332,
+      "learning_rate": 0.00019487870911092214,
+      "loss": 2.0326,
+      "step": 630
+    },
+    {
+      "epoch": 0.5977118841933224,
+      "grad_norm": 0.03215005545393897,
+      "learning_rate": 0.00019463178699506277,
+      "loss": 2.0231,
+      "step": 640
+    },
+    {
+      "epoch": 0.6070511323838431,
+      "grad_norm": 0.03823630791937631,
+      "learning_rate": 0.00019437921551104933,
+      "loss": 2.0293,
+      "step": 650
+    },
+    {
+      "epoch": 0.6163903805743638,
+      "grad_norm": 0.03200103149471209,
+      "learning_rate": 0.00019412100973700038,
+      "loss": 2.017,
+      "step": 660
+    },
+    {
+      "epoch": 0.6257296287648845,
+      "grad_norm": 0.03841804153577787,
+      "learning_rate": 0.00019385718508739262,
+      "loss": 2.0135,
+      "step": 670
+    },
+    {
+      "epoch": 0.635068876955405,
+      "grad_norm": 0.03052396655271533,
+      "learning_rate": 0.0001935877573121407,
+      "loss": 2.0237,
+      "step": 680
+    },
+    {
+      "epoch": 0.6444081251459257,
+      "grad_norm": 0.033817837533771815,
+      "learning_rate": 0.00019331274249565717,
+      "loss": 2.0069,
+      "step": 690
+    },
+    {
+      "epoch": 0.6537473733364464,
+      "grad_norm": 0.028286999650643876,
+      "learning_rate": 0.00019303215705589194,
+      "loss": 2.0112,
+      "step": 700
+    },
+    {
+      "epoch": 0.6630866215269671,
+      "grad_norm": 0.030693948626257357,
+      "learning_rate": 0.00019274601774335243,
+      "loss": 2.0,
+      "step": 710
+    },
+    {
+      "epoch": 0.6724258697174877,
+      "grad_norm": 0.029984615635953022,
+      "learning_rate": 0.0001924543416401035,
+      "loss": 2.0028,
+      "step": 720
+    },
+    {
+      "epoch": 0.6817651179080084,
+      "grad_norm": 0.02915985673921391,
+      "learning_rate": 0.00019215714615874755,
+      "loss": 2.0031,
+      "step": 730
+    },
+    {
+      "epoch": 0.6911043660985291,
+      "grad_norm": 0.028305399777245336,
+      "learning_rate": 0.00019185444904138528,
+      "loss": 1.9924,
+      "step": 740
+    },
+    {
+      "epoch": 0.7004436142890498,
+      "grad_norm": 0.036720505429756495,
+      "learning_rate": 0.00019154626835855628,
+      "loss": 1.9981,
+      "step": 750
+    },
+    {
+      "epoch": 0.7097828624795703,
+      "grad_norm": 0.0287124048917296,
+      "learning_rate": 0.00019123262250816034,
+      "loss": 1.9868,
+      "step": 760
+    },
+    {
+      "epoch": 0.719122110670091,
+      "grad_norm": 0.03318092492837997,
+      "learning_rate": 0.00019091353021435915,
+      "loss": 1.9943,
+      "step": 770
+    },
+    {
+      "epoch": 0.7284613588606117,
+      "grad_norm": 0.054035272137015325,
+      "learning_rate": 0.00019058901052645844,
+      "loss": 1.9838,
+      "step": 780
+    },
+    {
+      "epoch": 0.7378006070511324,
+      "grad_norm": 0.03184392761983255,
+      "learning_rate": 0.00019025908281777078,
+      "loss": 1.982,
+      "step": 790
+    },
+    {
+      "epoch": 0.747139855241653,
+      "grad_norm": 0.029409948164434735,
+      "learning_rate": 0.00018992376678445908,
+      "loss": 1.9693,
+      "step": 800
+    },
+    {
+      "epoch": 0.7564791034321737,
+      "grad_norm": 0.029656963043919016,
+      "learning_rate": 0.00018958308244436064,
+      "loss": 1.9914,
+      "step": 810
+    },
+    {
+      "epoch": 0.7658183516226944,
+      "grad_norm": 0.030843610865326686,
+      "learning_rate": 0.00018923705013579233,
+      "loss": 1.9749,
+      "step": 820
+    },
+    {
+      "epoch": 0.7751575998132151,
+      "grad_norm": 0.0377584286045999,
+      "learning_rate": 0.00018888569051633613,
+      "loss": 1.9606,
+      "step": 830
+    },
+    {
+      "epoch": 0.7844968480037356,
+      "grad_norm": 0.03147699916274391,
+      "learning_rate": 0.00018852902456160616,
+      "loss": 1.9696,
+      "step": 840
+    },
+    {
+      "epoch": 0.7938360961942563,
+      "grad_norm": 0.029193153251471263,
+      "learning_rate": 0.0001881670735639963,
+      "loss": 1.9687,
+      "step": 850
+    },
+    {
+      "epoch": 0.803175344384777,
+      "grad_norm": 0.030201895228156087,
+      "learning_rate": 0.00018779985913140924,
+      "loss": 1.9678,
+      "step": 860
+    },
+    {
+      "epoch": 0.8125145925752977,
+      "grad_norm": 0.029272319485493213,
+      "learning_rate": 0.00018742740318596632,
+      "loss": 1.9697,
+      "step": 870
+    },
+    {
+      "epoch": 0.8218538407658184,
+      "grad_norm": 0.033740176465285654,
+      "learning_rate": 0.000187049727962699,
+      "loss": 1.9647,
+      "step": 880
+    },
+    {
+      "epoch": 0.831193088956339,
+      "grad_norm": 0.029539399251208593,
+      "learning_rate": 0.0001866668560082213,
+      "loss": 1.9627,
+      "step": 890
+    },
+    {
+      "epoch": 0.8405323371468597,
+      "grad_norm": 0.028666077337951026,
+      "learning_rate": 0.0001862788101793839,
+      "loss": 1.9529,
+      "step": 900
+    },
+    {
+      "epoch": 0.8498715853373804,
+      "grad_norm": 0.030280792492665805,
+      "learning_rate": 0.0001858856136419097,
+      "loss": 1.9536,
+      "step": 910
+    },
+    {
+      "epoch": 0.859210833527901,
+      "grad_norm": 0.03921536105057096,
+      "learning_rate": 0.0001854872898690106,
+      "loss": 1.9474,
+      "step": 920
+    },
+    {
+      "epoch": 0.8685500817184216,
+      "grad_norm": 0.030632523637038354,
+      "learning_rate": 0.0001850838626399865,
+      "loss": 1.9423,
+      "step": 930
+    },
+    {
+      "epoch": 0.8778893299089423,
+      "grad_norm": 0.04615147601979514,
+      "learning_rate": 0.00018467535603880548,
+      "loss": 1.946,
+      "step": 940
+    },
+    {
+      "epoch": 0.887228578099463,
+      "grad_norm": 0.028216236017006333,
+      "learning_rate": 0.00018426179445266616,
+      "loss": 1.9408,
+      "step": 950
+    },
+    {
+      "epoch": 0.8965678262899837,
+      "grad_norm": 0.0282407563402959,
+      "learning_rate": 0.00018384320257054177,
+      "loss": 1.9447,
+      "step": 960
+    },
+    {
+      "epoch": 0.9059070744805043,
+      "grad_norm": 0.029365880854687894,
+      "learning_rate": 0.0001834196053817062,
+      "loss": 1.9389,
+      "step": 970
+    },
+    {
+      "epoch": 0.915246322671025,
+      "grad_norm": 0.02855404439130719,
+      "learning_rate": 0.00018299102817424234,
+      "loss": 1.9425,
+      "step": 980
+    },
+    {
+      "epoch": 0.9245855708615457,
+      "grad_norm": 0.03238310984070135,
+      "learning_rate": 0.00018255749653353225,
+      "loss": 1.9392,
+      "step": 990
+    },
+    {
+      "epoch": 0.9339248190520663,
+      "grad_norm": 0.030115350805430388,
+      "learning_rate": 0.00018211903634072983,
+      "loss": 1.956,
+      "step": 1000
+    },
+    {
+      "epoch": 0.9432640672425869,
+      "grad_norm": 0.039090564032501135,
+      "learning_rate": 0.0001816756737712158,
+      "loss": 1.9358,
+      "step": 1010
+    },
+    {
+      "epoch": 0.9526033154331076,
+      "grad_norm": 0.028567954476327023,
+      "learning_rate": 0.000181227435293035,
+      "loss": 1.9342,
+      "step": 1020
+    },
+    {
+      "epoch": 0.9619425636236283,
+      "grad_norm": 0.028158653241284505,
+      "learning_rate": 0.00018077434766531624,
+      "loss": 1.9287,
+      "step": 1030
+    },
+    {
+      "epoch": 0.971281811814149,
+      "grad_norm": 0.02927795717651538,
+      "learning_rate": 0.00018031643793667504,
+      "loss": 1.9298,
+      "step": 1040
+    },
+    {
+      "epoch": 0.9806210600046696,
+      "grad_norm": 0.06298340584032344,
+      "learning_rate": 0.0001798537334435986,
+      "loss": 1.9303,
+      "step": 1050
+    },
+    {
+      "epoch": 0.9899603081951903,
+      "grad_norm": 0.03627278406983255,
+      "learning_rate": 0.00017938626180881407,
+      "loss": 1.9285,
+      "step": 1060
+    },
+    {
+      "epoch": 0.999299556385711,
+      "grad_norm": 0.03916823421329747,
+      "learning_rate": 0.00017891405093963938,
+      "loss": 1.9239,
+      "step": 1070
+    },
+    {
+      "epoch": 1.0087555451786132,
+      "grad_norm": 0.03383539251115568,
+      "learning_rate": 0.00017843712902631723,
+      "loss": 1.8855,
+      "step": 1080
+    },
+    {
+      "epoch": 1.0180947933691338,
+      "grad_norm": 0.02931021936560147,
+      "learning_rate": 0.00017795552454033224,
+      "loss": 1.8004,
+      "step": 1090
+    },
+    {
+      "epoch": 1.0274340415596543,
+      "grad_norm": 0.029173295095350292,
+      "learning_rate": 0.0001774692662327113,
+      "loss": 1.7912,
+      "step": 1100
+    },
+    {
+      "epoch": 1.0367732897501751,
+      "grad_norm": 0.035210065642974735,
+      "learning_rate": 0.000176978383132307,
+      "loss": 1.7902,
+      "step": 1110
+    },
+    {
+      "epoch": 1.0461125379406957,
+      "grad_norm": 0.029794447659573477,
+      "learning_rate": 0.00017648290454406475,
+      "loss": 1.8072,
+      "step": 1120
+    },
+    {
+      "epoch": 1.0554517861312165,
+      "grad_norm": 0.03194584938279939,
+      "learning_rate": 0.0001759828600472734,
+      "loss": 1.803,
+      "step": 1130
+    },
+    {
+      "epoch": 1.064791034321737,
+      "grad_norm": 0.031028415312581603,
+      "learning_rate": 0.00017547827949379924,
+      "loss": 1.7945,
+      "step": 1140
+    },
+    {
+      "epoch": 1.0741302825122578,
+      "grad_norm": 0.03313245828751237,
+      "learning_rate": 0.00017496919300630403,
+      "loss": 1.8139,
+      "step": 1150
+    },
+    {
+      "epoch": 1.0834695307027784,
+      "grad_norm": 0.03049431724979126,
+      "learning_rate": 0.00017445563097644664,
+      "loss": 1.8031,
+      "step": 1160
+    },
+    {
+      "epoch": 1.0928087788932992,
+      "grad_norm": 0.02880548705343715,
+      "learning_rate": 0.00017393762406306878,
+      "loss": 1.7974,
+      "step": 1170
+    },
+    {
+      "epoch": 1.1021480270838198,
+      "grad_norm": 0.03200427895977668,
+      "learning_rate": 0.00017341520319036469,
+      "loss": 1.7994,
+      "step": 1180
+    },
+    {
+      "epoch": 1.1114872752743403,
+      "grad_norm": 0.031293532815600045,
+      "learning_rate": 0.00017288839954603496,
+      "loss": 1.8073,
+      "step": 1190
+    },
+    {
+      "epoch": 1.1208265234648611,
+      "grad_norm": 0.032367211109345505,
+      "learning_rate": 0.00017235724457942468,
+      "loss": 1.7944,
+      "step": 1200
+    },
+    {
+      "epoch": 1.1301657716553817,
+      "grad_norm": 0.037783793923191374,
+      "learning_rate": 0.0001718217699996462,
+      "loss": 1.7948,
+      "step": 1210
+    },
+    {
+      "epoch": 1.1395050198459025,
+      "grad_norm": 0.02881083117349317,
+      "learning_rate": 0.00017128200777368567,
+      "loss": 1.8029,
+      "step": 1220
+    },
+    {
+      "epoch": 1.148844268036423,
+      "grad_norm": 0.032997408408175985,
+      "learning_rate": 0.00017073799012449524,
+      "loss": 1.7914,
+      "step": 1230
+    },
+    {
+      "epoch": 1.1581835162269438,
+      "grad_norm": 0.030263615801190885,
+      "learning_rate": 0.00017018974952906884,
+      "loss": 1.792,
+      "step": 1240
+    },
+    {
+      "epoch": 1.1675227644174644,
+      "grad_norm": 0.03286310429098484,
+      "learning_rate": 0.00016963731871650378,
+      "loss": 1.8149,
+      "step": 1250
+    },
+    {
+      "epoch": 1.1768620126079852,
+      "grad_norm": 0.031111621587597126,
+      "learning_rate": 0.00016908073066604663,
+      "loss": 1.8092,
+      "step": 1260
+    },
+    {
+      "epoch": 1.1862012607985057,
+      "grad_norm": 0.03372950172075241,
+      "learning_rate": 0.0001685200186051246,
+      "loss": 1.818,
+      "step": 1270
+    },
+    {
+      "epoch": 1.1955405089890263,
+      "grad_norm": 0.040782040436567434,
+      "learning_rate": 0.00016795521600736164,
+      "loss": 1.7999,
+      "step": 1280
+    },
+    {
+      "epoch": 1.204879757179547,
+      "grad_norm": 0.03451605168178924,
+      "learning_rate": 0.00016738635659058044,
+      "loss": 1.7945,
+      "step": 1290
+    },
+    {
+      "epoch": 1.2142190053700677,
+      "grad_norm": 0.03235681588882673,
+      "learning_rate": 0.00016681347431478933,
+      "loss": 1.8087,
+      "step": 1300
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 4280,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 100,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.0526118085398626e+20,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

l2-13b-ga/checkpoint-1300/zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

l2-13b-ga/checkpoint-1600/model.safetensors.index.json

@@ -0,0 +1,370 @@
+{
+  "metadata": {
+    "total_size": 26103060480
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00006-of-00006.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.15.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.30.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.31.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.input_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.down_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.gate_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.up_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.post_attention_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.39.input_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.down_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.gate_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.up_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.post_attention_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.k_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.o_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.q_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.v_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.norm.weight": "model-00006-of-00006.safetensors"
+  }
+}

l2-13b-ga/checkpoint-1600/trainer_state.json

@@ -0,0 +1,1160 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.4943964510856875,
+  "eval_steps": 500,
+  "global_step": 1600,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0009339248190520663,
+      "grad_norm": 6.638877692627699,
+      "learning_rate": 9.345794392523364e-07,
+      "loss": 9.2917,
+      "step": 1
+    },
+    {
+      "epoch": 0.009339248190520663,
+      "grad_norm": 1.1560921335705272,
+      "learning_rate": 9.345794392523365e-06,
+      "loss": 9.0876,
+      "step": 10
+    },
+    {
+      "epoch": 0.018678496381041326,
+      "grad_norm": 0.8415132296956432,
+      "learning_rate": 1.869158878504673e-05,
+      "loss": 8.2164,
+      "step": 20
+    },
+    {
+      "epoch": 0.02801774457156199,
+      "grad_norm": 0.45381630992958155,
+      "learning_rate": 2.8037383177570094e-05,
+      "loss": 7.5184,
+      "step": 30
+    },
+    {
+      "epoch": 0.03735699276208265,
+      "grad_norm": 0.8400636107958425,
+      "learning_rate": 3.738317757009346e-05,
+      "loss": 6.6507,
+      "step": 40
+    },
+    {
+      "epoch": 0.046696240952603316,
+      "grad_norm": 0.557696240829066,
+      "learning_rate": 4.672897196261683e-05,
+      "loss": 5.8909,
+      "step": 50
+    },
+    {
+      "epoch": 0.05603548914312398,
+      "grad_norm": 0.3971996057467842,
+      "learning_rate": 5.607476635514019e-05,
+      "loss": 5.4127,
+      "step": 60
+    },
+    {
+      "epoch": 0.06537473733364464,
+      "grad_norm": 0.2932710540265688,
+      "learning_rate": 6.542056074766355e-05,
+      "loss": 5.0106,
+      "step": 70
+    },
+    {
+      "epoch": 0.0747139855241653,
+      "grad_norm": 0.3682690443551033,
+      "learning_rate": 7.476635514018692e-05,
+      "loss": 4.6042,
+      "step": 80
+    },
+    {
+      "epoch": 0.08405323371468597,
+      "grad_norm": 0.3132971920011515,
+      "learning_rate": 8.411214953271028e-05,
+      "loss": 4.2031,
+      "step": 90
+    },
+    {
+      "epoch": 0.09339248190520663,
+      "grad_norm": 0.6731868159213446,
+      "learning_rate": 9.345794392523365e-05,
+      "loss": 3.9423,
+      "step": 100
+    },
+    {
+      "epoch": 0.1027317300957273,
+      "grad_norm": 0.27848867836763197,
+      "learning_rate": 0.000102803738317757,
+      "loss": 3.7157,
+      "step": 110
+    },
+    {
+      "epoch": 0.11207097828624796,
+      "grad_norm": 0.24642109032991807,
+      "learning_rate": 0.00011214953271028037,
+      "loss": 3.4516,
+      "step": 120
+    },
+    {
+      "epoch": 0.12141022647676862,
+      "grad_norm": 0.25717384664029797,
+      "learning_rate": 0.00012149532710280373,
+      "loss": 3.2167,
+      "step": 130
+    },
+    {
+      "epoch": 0.13074947466728928,
+      "grad_norm": 0.20912922668565637,
+      "learning_rate": 0.0001308411214953271,
+      "loss": 3.0237,
+      "step": 140
+    },
+    {
+      "epoch": 0.14008872285780993,
+      "grad_norm": 0.15805888388706113,
+      "learning_rate": 0.00014018691588785047,
+      "loss": 2.8529,
+      "step": 150
+    },
+    {
+      "epoch": 0.1494279710483306,
+      "grad_norm": 0.23370349497479534,
+      "learning_rate": 0.00014953271028037384,
+      "loss": 2.7078,
+      "step": 160
+    },
+    {
+      "epoch": 0.15876721923885126,
+      "grad_norm": 0.1802138633012483,
+      "learning_rate": 0.0001588785046728972,
+      "loss": 2.6115,
+      "step": 170
+    },
+    {
+      "epoch": 0.16810646742937194,
+      "grad_norm": 0.13354347610039718,
+      "learning_rate": 0.00016822429906542056,
+      "loss": 2.5309,
+      "step": 180
+    },
+    {
+      "epoch": 0.17744571561989259,
+      "grad_norm": 0.09414865188086892,
+      "learning_rate": 0.00017757009345794393,
+      "loss": 2.4452,
+      "step": 190
+    },
+    {
+      "epoch": 0.18678496381041326,
+      "grad_norm": 0.08333601554768896,
+      "learning_rate": 0.0001869158878504673,
+      "loss": 2.3832,
+      "step": 200
+    },
+    {
+      "epoch": 0.1961242120009339,
+      "grad_norm": 0.15926414699806835,
+      "learning_rate": 0.00019626168224299065,
+      "loss": 2.3492,
+      "step": 210
+    },
+    {
+      "epoch": 0.2054634601914546,
+      "grad_norm": 0.09492820761057012,
+      "learning_rate": 0.0001999989254250208,
+      "loss": 2.323,
+      "step": 220
+    },
+    {
+      "epoch": 0.21480270838197524,
+      "grad_norm": 0.0801349259356147,
+      "learning_rate": 0.00019999235866155886,
+      "loss": 2.2731,
+      "step": 230
+    },
+    {
+      "epoch": 0.22414195657249592,
+      "grad_norm": 0.12210960524693895,
+      "learning_rate": 0.00019997982251228469,
+      "loss": 2.2433,
+      "step": 240
+    },
+    {
+      "epoch": 0.23348120476301656,
+      "grad_norm": 3.14289498732125,
+      "learning_rate": 0.00019996131772558666,
+      "loss": 3.2769,
+      "step": 250
+    },
+    {
+      "epoch": 0.24282045295353724,
+      "grad_norm": 1.632940983166179,
+      "learning_rate": 0.00019993684540617132,
+      "loss": 4.9343,
+      "step": 260
+    },
+    {
+      "epoch": 0.2521597011440579,
+      "grad_norm": 3.4831252230225416,
+      "learning_rate": 0.00019990640701499736,
+      "loss": 4.2768,
+      "step": 270
+    },
+    {
+      "epoch": 0.26149894933457857,
+      "grad_norm": 1.6069045920523788,
+      "learning_rate": 0.00019987000436918874,
+      "loss": 5.9581,
+      "step": 280
+    },
+    {
+      "epoch": 0.27083819752509924,
+      "grad_norm": 0.2220907936615993,
+      "learning_rate": 0.00019982763964192585,
+      "loss": 3.8228,
+      "step": 290
+    },
+    {
+      "epoch": 0.28017744571561987,
+      "grad_norm": 0.24737284913291765,
+      "learning_rate": 0.00019977931536231596,
+      "loss": 3.1413,
+      "step": 300
+    },
+    {
+      "epoch": 0.28951669390614054,
+      "grad_norm": 4.010404518241152,
+      "learning_rate": 0.00019972503441524224,
+      "loss": 2.8432,
+      "step": 310
+    },
+    {
+      "epoch": 0.2988559420966612,
+      "grad_norm": 0.1515583580811596,
+      "learning_rate": 0.00019966480004119142,
+      "loss": 2.7859,
+      "step": 320
+    },
+    {
+      "epoch": 0.3081951902871819,
+      "grad_norm": 0.11259395750650594,
+      "learning_rate": 0.00019959861583606045,
+      "loss": 2.5821,
+      "step": 330
+    },
+    {
+      "epoch": 0.3175344384777025,
+      "grad_norm": 0.22514797814956813,
+      "learning_rate": 0.00019952648575094183,
+      "loss": 2.4517,
+      "step": 340
+    },
+    {
+      "epoch": 0.3268736866682232,
+      "grad_norm": 0.08040136172033542,
+      "learning_rate": 0.00019944841409188767,
+      "loss": 2.3794,
+      "step": 350
+    },
+    {
+      "epoch": 0.3362129348587439,
+      "grad_norm": 0.054758073593565354,
+      "learning_rate": 0.00019936440551965263,
+      "loss": 2.3232,
+      "step": 360
+    },
+    {
+      "epoch": 0.34555218304926455,
+      "grad_norm": 0.06742998909645591,
+      "learning_rate": 0.00019927446504941577,
+      "loss": 2.2776,
+      "step": 370
+    },
+    {
+      "epoch": 0.35489143123978517,
+      "grad_norm": 0.048780907584876736,
+      "learning_rate": 0.00019917859805048096,
+      "loss": 2.2376,
+      "step": 380
+    },
+    {
+      "epoch": 0.36423067943030585,
+      "grad_norm": 0.0475325963052214,
+      "learning_rate": 0.00019907681024595663,
+      "loss": 2.2191,
+      "step": 390
+    },
+    {
+      "epoch": 0.3735699276208265,
+      "grad_norm": 0.054089563211590065,
+      "learning_rate": 0.00019896910771241387,
+      "loss": 2.1961,
+      "step": 400
+    },
+    {
+      "epoch": 0.3829091758113472,
+      "grad_norm": 0.21798406131864823,
+      "learning_rate": 0.00019885549687952372,
+      "loss": 2.2078,
+      "step": 410
+    },
+    {
+      "epoch": 0.3922484240018678,
+      "grad_norm": 0.8673185709111124,
+      "learning_rate": 0.00019873598452967338,
+      "loss": 2.3731,
+      "step": 420
+    },
+    {
+      "epoch": 0.4015876721923885,
+      "grad_norm": 0.22424350669971718,
+      "learning_rate": 0.0001986105777975613,
+      "loss": 2.6195,
+      "step": 430
+    },
+    {
+      "epoch": 0.4109269203829092,
+      "grad_norm": 0.307418135168262,
+      "learning_rate": 0.00019847928416977126,
+      "loss": 2.3624,
+      "step": 440
+    },
+    {
+      "epoch": 0.42026616857342985,
+      "grad_norm": 0.07944722668080402,
+      "learning_rate": 0.00019834211148432536,
+      "loss": 2.2799,
+      "step": 450
+    },
+    {
+      "epoch": 0.4296054167639505,
+      "grad_norm": 0.18146933758664588,
+      "learning_rate": 0.00019819906793021614,
+      "loss": 2.2177,
+      "step": 460
+    },
+    {
+      "epoch": 0.43894466495447115,
+      "grad_norm": 0.07035825837333018,
+      "learning_rate": 0.0001980501620469178,
+      "loss": 2.1767,
+      "step": 470
+    },
+    {
+      "epoch": 0.44828391314499183,
+      "grad_norm": 0.04596186944454228,
+      "learning_rate": 0.0001978954027238763,
+      "loss": 2.1598,
+      "step": 480
+    },
+    {
+      "epoch": 0.4576231613355125,
+      "grad_norm": 0.041342347745088055,
+      "learning_rate": 0.0001977347991999786,
+      "loss": 2.131,
+      "step": 490
+    },
+    {
+      "epoch": 0.46696240952603313,
+      "grad_norm": 0.04172063219841485,
+      "learning_rate": 0.00019756836106300137,
+      "loss": 2.1231,
+      "step": 500
+    },
+    {
+      "epoch": 0.4763016577165538,
+      "grad_norm": 0.03373646457711144,
+      "learning_rate": 0.00019739609824903843,
+      "loss": 2.1146,
+      "step": 510
+    },
+    {
+      "epoch": 0.4856409059070745,
+      "grad_norm": 0.03736871030676605,
+      "learning_rate": 0.00019721802104190748,
+      "loss": 2.1003,
+      "step": 520
+    },
+    {
+      "epoch": 0.49498015409759516,
+      "grad_norm": 0.033931028038211034,
+      "learning_rate": 0.00019703414007253645,
+      "loss": 2.0983,
+      "step": 530
+    },
+    {
+      "epoch": 0.5043194022881158,
+      "grad_norm": 0.03790055446070549,
+      "learning_rate": 0.00019684446631832868,
+      "loss": 2.092,
+      "step": 540
+    },
+    {
+      "epoch": 0.5136586504786365,
+      "grad_norm": 0.030956192803893078,
+      "learning_rate": 0.00019664901110250758,
+      "loss": 2.0807,
+      "step": 550
+    },
+    {
+      "epoch": 0.5229978986691571,
+      "grad_norm": 0.03542530209935129,
+      "learning_rate": 0.00019644778609344068,
+      "loss": 2.0773,
+      "step": 560
+    },
+    {
+      "epoch": 0.5323371468596778,
+      "grad_norm": 0.040947757568902336,
+      "learning_rate": 0.00019624080330394306,
+      "loss": 2.0649,
+      "step": 570
+    },
+    {
+      "epoch": 0.5416763950501985,
+      "grad_norm": 0.034273415973688146,
+      "learning_rate": 0.00019602807509056018,
+      "loss": 2.0479,
+      "step": 580
+    },
+    {
+      "epoch": 0.5510156432407192,
+      "grad_norm": 0.031427481498873144,
+      "learning_rate": 0.00019580961415283028,
+      "loss": 2.0563,
+      "step": 590
+    },
+    {
+      "epoch": 0.5603548914312397,
+      "grad_norm": 0.03141549752041532,
+      "learning_rate": 0.00019558543353252611,
+      "loss": 2.0503,
+      "step": 600
+    },
+    {
+      "epoch": 0.5696941396217604,
+      "grad_norm": 0.033012392726428204,
+      "learning_rate": 0.00019535554661287652,
+      "loss": 2.0389,
+      "step": 610
+    },
+    {
+      "epoch": 0.5790333878122811,
+      "grad_norm": 0.02913261992661444,
+      "learning_rate": 0.0001951199671177673,
+      "loss": 2.036,
+      "step": 620
+    },
+    {
+      "epoch": 0.5883726360028018,
+      "grad_norm": 0.030543903708435332,
+      "learning_rate": 0.00019487870911092214,
+      "loss": 2.0326,
+      "step": 630
+    },
+    {
+      "epoch": 0.5977118841933224,
+      "grad_norm": 0.03215005545393897,
+      "learning_rate": 0.00019463178699506277,
+      "loss": 2.0231,
+      "step": 640
+    },
+    {
+      "epoch": 0.6070511323838431,
+      "grad_norm": 0.03823630791937631,
+      "learning_rate": 0.00019437921551104933,
+      "loss": 2.0293,
+      "step": 650
+    },
+    {
+      "epoch": 0.6163903805743638,
+      "grad_norm": 0.03200103149471209,
+      "learning_rate": 0.00019412100973700038,
+      "loss": 2.017,
+      "step": 660
+    },
+    {
+      "epoch": 0.6257296287648845,
+      "grad_norm": 0.03841804153577787,
+      "learning_rate": 0.00019385718508739262,
+      "loss": 2.0135,
+      "step": 670
+    },
+    {
+      "epoch": 0.635068876955405,
+      "grad_norm": 0.03052396655271533,
+      "learning_rate": 0.0001935877573121407,
+      "loss": 2.0237,
+      "step": 680
+    },
+    {
+      "epoch": 0.6444081251459257,
+      "grad_norm": 0.033817837533771815,
+      "learning_rate": 0.00019331274249565717,
+      "loss": 2.0069,
+      "step": 690
+    },
+    {
+      "epoch": 0.6537473733364464,
+      "grad_norm": 0.028286999650643876,
+      "learning_rate": 0.00019303215705589194,
+      "loss": 2.0112,
+      "step": 700
+    },
+    {
+      "epoch": 0.6630866215269671,
+      "grad_norm": 0.030693948626257357,
+      "learning_rate": 0.00019274601774335243,
+      "loss": 2.0,
+      "step": 710
+    },
+    {
+      "epoch": 0.6724258697174877,
+      "grad_norm": 0.029984615635953022,
+      "learning_rate": 0.0001924543416401035,
+      "loss": 2.0028,
+      "step": 720
+    },
+    {
+      "epoch": 0.6817651179080084,
+      "grad_norm": 0.02915985673921391,
+      "learning_rate": 0.00019215714615874755,
+      "loss": 2.0031,
+      "step": 730
+    },
+    {
+      "epoch": 0.6911043660985291,
+      "grad_norm": 0.028305399777245336,
+      "learning_rate": 0.00019185444904138528,
+      "loss": 1.9924,
+      "step": 740
+    },
+    {
+      "epoch": 0.7004436142890498,
+      "grad_norm": 0.036720505429756495,
+      "learning_rate": 0.00019154626835855628,
+      "loss": 1.9981,
+      "step": 750
+    },
+    {
+      "epoch": 0.7097828624795703,
+      "grad_norm": 0.0287124048917296,
+      "learning_rate": 0.00019123262250816034,
+      "loss": 1.9868,
+      "step": 760
+    },
+    {
+      "epoch": 0.719122110670091,
+      "grad_norm": 0.03318092492837997,
+      "learning_rate": 0.00019091353021435915,
+      "loss": 1.9943,
+      "step": 770
+    },
+    {
+      "epoch": 0.7284613588606117,
+      "grad_norm": 0.054035272137015325,
+      "learning_rate": 0.00019058901052645844,
+      "loss": 1.9838,
+      "step": 780
+    },
+    {
+      "epoch": 0.7378006070511324,
+      "grad_norm": 0.03184392761983255,
+      "learning_rate": 0.00019025908281777078,
+      "loss": 1.982,
+      "step": 790
+    },
+    {
+      "epoch": 0.747139855241653,
+      "grad_norm": 0.029409948164434735,
+      "learning_rate": 0.00018992376678445908,
+      "loss": 1.9693,
+      "step": 800
+    },
+    {
+      "epoch": 0.7564791034321737,
+      "grad_norm": 0.029656963043919016,
+      "learning_rate": 0.00018958308244436064,
+      "loss": 1.9914,
+      "step": 810
+    },
+    {
+      "epoch": 0.7658183516226944,
+      "grad_norm": 0.030843610865326686,
+      "learning_rate": 0.00018923705013579233,
+      "loss": 1.9749,
+      "step": 820
+    },
+    {
+      "epoch": 0.7751575998132151,
+      "grad_norm": 0.0377584286045999,
+      "learning_rate": 0.00018888569051633613,
+      "loss": 1.9606,
+      "step": 830
+    },
+    {
+      "epoch": 0.7844968480037356,
+      "grad_norm": 0.03147699916274391,
+      "learning_rate": 0.00018852902456160616,
+      "loss": 1.9696,
+      "step": 840
+    },
+    {
+      "epoch": 0.7938360961942563,
+      "grad_norm": 0.029193153251471263,
+      "learning_rate": 0.0001881670735639963,
+      "loss": 1.9687,
+      "step": 850
+    },
+    {
+      "epoch": 0.803175344384777,
+      "grad_norm": 0.030201895228156087,
+      "learning_rate": 0.00018779985913140924,
+      "loss": 1.9678,
+      "step": 860
+    },
+    {
+      "epoch": 0.8125145925752977,
+      "grad_norm": 0.029272319485493213,
+      "learning_rate": 0.00018742740318596632,
+      "loss": 1.9697,
+      "step": 870
+    },
+    {
+      "epoch": 0.8218538407658184,
+      "grad_norm": 0.033740176465285654,
+      "learning_rate": 0.000187049727962699,
+      "loss": 1.9647,
+      "step": 880
+    },
+    {
+      "epoch": 0.831193088956339,
+      "grad_norm": 0.029539399251208593,
+      "learning_rate": 0.0001866668560082213,
+      "loss": 1.9627,
+      "step": 890
+    },
+    {
+      "epoch": 0.8405323371468597,
+      "grad_norm": 0.028666077337951026,
+      "learning_rate": 0.0001862788101793839,
+      "loss": 1.9529,
+      "step": 900
+    },
+    {
+      "epoch": 0.8498715853373804,
+      "grad_norm": 0.030280792492665805,
+      "learning_rate": 0.0001858856136419097,
+      "loss": 1.9536,
+      "step": 910
+    },
+    {
+      "epoch": 0.859210833527901,
+      "grad_norm": 0.03921536105057096,
+      "learning_rate": 0.0001854872898690106,
+      "loss": 1.9474,
+      "step": 920
+    },
+    {
+      "epoch": 0.8685500817184216,
+      "grad_norm": 0.030632523637038354,
+      "learning_rate": 0.0001850838626399865,
+      "loss": 1.9423,
+      "step": 930
+    },
+    {
+      "epoch": 0.8778893299089423,
+      "grad_norm": 0.04615147601979514,
+      "learning_rate": 0.00018467535603880548,
+      "loss": 1.946,
+      "step": 940
+    },
+    {
+      "epoch": 0.887228578099463,
+      "grad_norm": 0.028216236017006333,
+      "learning_rate": 0.00018426179445266616,
+      "loss": 1.9408,
+      "step": 950
+    },
+    {
+      "epoch": 0.8965678262899837,
+      "grad_norm": 0.0282407563402959,
+      "learning_rate": 0.00018384320257054177,
+      "loss": 1.9447,
+      "step": 960
+    },
+    {
+      "epoch": 0.9059070744805043,
+      "grad_norm": 0.029365880854687894,
+      "learning_rate": 0.0001834196053817062,
+      "loss": 1.9389,
+      "step": 970
+    },
+    {
+      "epoch": 0.915246322671025,
+      "grad_norm": 0.02855404439130719,
+      "learning_rate": 0.00018299102817424234,
+      "loss": 1.9425,
+      "step": 980
+    },
+    {
+      "epoch": 0.9245855708615457,
+      "grad_norm": 0.03238310984070135,
+      "learning_rate": 0.00018255749653353225,
+      "loss": 1.9392,
+      "step": 990
+    },
+    {
+      "epoch": 0.9339248190520663,
+      "grad_norm": 0.030115350805430388,
+      "learning_rate": 0.00018211903634072983,
+      "loss": 1.956,
+      "step": 1000
+    },
+    {
+      "epoch": 0.9432640672425869,
+      "grad_norm": 0.039090564032501135,
+      "learning_rate": 0.0001816756737712158,
+      "loss": 1.9358,
+      "step": 1010
+    },
+    {
+      "epoch": 0.9526033154331076,
+      "grad_norm": 0.028567954476327023,
+      "learning_rate": 0.000181227435293035,
+      "loss": 1.9342,
+      "step": 1020
+    },
+    {
+      "epoch": 0.9619425636236283,
+      "grad_norm": 0.028158653241284505,
+      "learning_rate": 0.00018077434766531624,
+      "loss": 1.9287,
+      "step": 1030
+    },
+    {
+      "epoch": 0.971281811814149,
+      "grad_norm": 0.02927795717651538,
+      "learning_rate": 0.00018031643793667504,
+      "loss": 1.9298,
+      "step": 1040
+    },
+    {
+      "epoch": 0.9806210600046696,
+      "grad_norm": 0.06298340584032344,
+      "learning_rate": 0.0001798537334435986,
+      "loss": 1.9303,
+      "step": 1050
+    },
+    {
+      "epoch": 0.9899603081951903,
+      "grad_norm": 0.03627278406983255,
+      "learning_rate": 0.00017938626180881407,
+      "loss": 1.9285,
+      "step": 1060
+    },
+    {
+      "epoch": 0.999299556385711,
+      "grad_norm": 0.03916823421329747,
+      "learning_rate": 0.00017891405093963938,
+      "loss": 1.9239,
+      "step": 1070
+    },
+    {
+      "epoch": 1.0087555451786132,
+      "grad_norm": 0.03383539251115568,
+      "learning_rate": 0.00017843712902631723,
+      "loss": 1.8855,
+      "step": 1080
+    },
+    {
+      "epoch": 1.0180947933691338,
+      "grad_norm": 0.02931021936560147,
+      "learning_rate": 0.00017795552454033224,
+      "loss": 1.8004,
+      "step": 1090
+    },
+    {
+      "epoch": 1.0274340415596543,
+      "grad_norm": 0.029173295095350292,
+      "learning_rate": 0.0001774692662327113,
+      "loss": 1.7912,
+      "step": 1100
+    },
+    {
+      "epoch": 1.0367732897501751,
+      "grad_norm": 0.035210065642974735,
+      "learning_rate": 0.000176978383132307,
+      "loss": 1.7902,
+      "step": 1110
+    },
+    {
+      "epoch": 1.0461125379406957,
+      "grad_norm": 0.029794447659573477,
+      "learning_rate": 0.00017648290454406475,
+      "loss": 1.8072,
+      "step": 1120
+    },
+    {
+      "epoch": 1.0554517861312165,
+      "grad_norm": 0.03194584938279939,
+      "learning_rate": 0.0001759828600472734,
+      "loss": 1.803,
+      "step": 1130
+    },
+    {
+      "epoch": 1.064791034321737,
+      "grad_norm": 0.031028415312581603,
+      "learning_rate": 0.00017547827949379924,
+      "loss": 1.7945,
+      "step": 1140
+    },
+    {
+      "epoch": 1.0741302825122578,
+      "grad_norm": 0.03313245828751237,
+      "learning_rate": 0.00017496919300630403,
+      "loss": 1.8139,
+      "step": 1150
+    },
+    {
+      "epoch": 1.0834695307027784,
+      "grad_norm": 0.03049431724979126,
+      "learning_rate": 0.00017445563097644664,
+      "loss": 1.8031,
+      "step": 1160
+    },
+    {
+      "epoch": 1.0928087788932992,
+      "grad_norm": 0.02880548705343715,
+      "learning_rate": 0.00017393762406306878,
+      "loss": 1.7974,
+      "step": 1170
+    },
+    {
+      "epoch": 1.1021480270838198,
+      "grad_norm": 0.03200427895977668,
+      "learning_rate": 0.00017341520319036469,
+      "loss": 1.7994,
+      "step": 1180
+    },
+    {
+      "epoch": 1.1114872752743403,
+      "grad_norm": 0.031293532815600045,
+      "learning_rate": 0.00017288839954603496,
+      "loss": 1.8073,
+      "step": 1190
+    },
+    {
+      "epoch": 1.1208265234648611,
+      "grad_norm": 0.032367211109345505,
+      "learning_rate": 0.00017235724457942468,
+      "loss": 1.7944,
+      "step": 1200
+    },
+    {
+      "epoch": 1.1301657716553817,
+      "grad_norm": 0.037783793923191374,
+      "learning_rate": 0.0001718217699996462,
+      "loss": 1.7948,
+      "step": 1210
+    },
+    {
+      "epoch": 1.1395050198459025,
+      "grad_norm": 0.02881083117349317,
+      "learning_rate": 0.00017128200777368567,
+      "loss": 1.8029,
+      "step": 1220
+    },
+    {
+      "epoch": 1.148844268036423,
+      "grad_norm": 0.032997408408175985,
+      "learning_rate": 0.00017073799012449524,
+      "loss": 1.7914,
+      "step": 1230
+    },
+    {
+      "epoch": 1.1581835162269438,
+      "grad_norm": 0.030263615801190885,
+      "learning_rate": 0.00017018974952906884,
+      "loss": 1.792,
+      "step": 1240
+    },
+    {
+      "epoch": 1.1675227644174644,
+      "grad_norm": 0.03286310429098484,
+      "learning_rate": 0.00016963731871650378,
+      "loss": 1.8149,
+      "step": 1250
+    },
+    {
+      "epoch": 1.1768620126079852,
+      "grad_norm": 0.031111621587597126,
+      "learning_rate": 0.00016908073066604663,
+      "loss": 1.8092,
+      "step": 1260
+    },
+    {
+      "epoch": 1.1862012607985057,
+      "grad_norm": 0.03372950172075241,
+      "learning_rate": 0.0001685200186051246,
+      "loss": 1.818,
+      "step": 1270
+    },
+    {
+      "epoch": 1.1955405089890263,
+      "grad_norm": 0.040782040436567434,
+      "learning_rate": 0.00016795521600736164,
+      "loss": 1.7999,
+      "step": 1280
+    },
+    {
+      "epoch": 1.204879757179547,
+      "grad_norm": 0.03451605168178924,
+      "learning_rate": 0.00016738635659058044,
+      "loss": 1.7945,
+      "step": 1290
+    },
+    {
+      "epoch": 1.2142190053700677,
+      "grad_norm": 0.03235681588882673,
+      "learning_rate": 0.00016681347431478933,
+      "loss": 1.8087,
+      "step": 1300
+    },
+    {
+      "epoch": 1.2235582535605884,
+      "grad_norm": 0.030750745605971932,
+      "learning_rate": 0.00016623660338015487,
+      "loss": 1.7995,
+      "step": 1310
+    },
+    {
+      "epoch": 1.232897501751109,
+      "grad_norm": 0.029444668665577274,
+      "learning_rate": 0.00016565577822496042,
+      "loss": 1.8025,
+      "step": 1320
+    },
+    {
+      "epoch": 1.2422367499416298,
+      "grad_norm": 0.038528856709584745,
+      "learning_rate": 0.00016507103352354996,
+      "loss": 1.7954,
+      "step": 1330
+    },
+    {
+      "epoch": 1.2515759981321504,
+      "grad_norm": 0.034217088004383035,
+      "learning_rate": 0.00016448240418425814,
+      "loss": 1.7962,
+      "step": 1340
+    },
+    {
+      "epoch": 1.260915246322671,
+      "grad_norm": 0.030205405393195585,
+      "learning_rate": 0.00016388992534732645,
+      "loss": 1.7973,
+      "step": 1350
+    },
+    {
+      "epoch": 1.2702544945131917,
+      "grad_norm": 0.029082218516562994,
+      "learning_rate": 0.00016329363238280528,
+      "loss": 1.796,
+      "step": 1360
+    },
+    {
+      "epoch": 1.2795937427037123,
+      "grad_norm": 0.029003887688766505,
+      "learning_rate": 0.00016269356088844238,
+      "loss": 1.7946,
+      "step": 1370
+    },
+    {
+      "epoch": 1.288932990894233,
+      "grad_norm": 0.03341157363649238,
+      "learning_rate": 0.00016208974668755779,
+      "loss": 1.7972,
+      "step": 1380
+    },
+    {
+      "epoch": 1.2982722390847536,
+      "grad_norm": 0.030614480844663026,
+      "learning_rate": 0.00016148222582690517,
+      "loss": 1.7973,
+      "step": 1390
+    },
+    {
+      "epoch": 1.3076114872752744,
+      "grad_norm": 0.029741346740467405,
+      "learning_rate": 0.00016087103457452,
+      "loss": 1.8076,
+      "step": 1400
+    },
+    {
+      "epoch": 1.316950735465795,
+      "grad_norm": 0.029569313554185597,
+      "learning_rate": 0.00016025620941755424,
+      "loss": 1.8043,
+      "step": 1410
+    },
+    {
+      "epoch": 1.3262899836563156,
+      "grad_norm": 0.02947637404374054,
+      "learning_rate": 0.0001596377870600983,
+      "loss": 1.797,
+      "step": 1420
+    },
+    {
+      "epoch": 1.3356292318468364,
+      "grad_norm": 0.031005062093959545,
+      "learning_rate": 0.00015901580442098968,
+      "loss": 1.8086,
+      "step": 1430
+    },
+    {
+      "epoch": 1.344968480037357,
+      "grad_norm": 0.029493792984873927,
+      "learning_rate": 0.00015839029863160922,
+      "loss": 1.8026,
+      "step": 1440
+    },
+    {
+      "epoch": 1.3543077282278777,
+      "grad_norm": 0.0288068155951218,
+      "learning_rate": 0.0001577613070336641,
+      "loss": 1.7951,
+      "step": 1450
+    },
+    {
+      "epoch": 1.3636469764183983,
+      "grad_norm": 0.03380404824627639,
+      "learning_rate": 0.00015712886717695885,
+      "loss": 1.7938,
+      "step": 1460
+    },
+    {
+      "epoch": 1.372986224608919,
+      "grad_norm": 0.039744706189693335,
+      "learning_rate": 0.0001564930168171536,
+      "loss": 1.8016,
+      "step": 1470
+    },
+    {
+      "epoch": 1.3823254727994396,
+      "grad_norm": 0.030565530594285437,
+      "learning_rate": 0.00015585379391351012,
+      "loss": 1.7984,
+      "step": 1480
+    },
+    {
+      "epoch": 1.3916647209899602,
+      "grad_norm": 0.04009392805554255,
+      "learning_rate": 0.00015521123662662567,
+      "loss": 1.7999,
+      "step": 1490
+    },
+    {
+      "epoch": 1.401003969180481,
+      "grad_norm": 0.03516196009586836,
+      "learning_rate": 0.000154565383316155,
+      "loss": 1.7979,
+      "step": 1500
+    },
+    {
+      "epoch": 1.4103432173710018,
+      "grad_norm": 0.03534161399054556,
+      "learning_rate": 0.0001539162725385202,
+      "loss": 1.8057,
+      "step": 1510
+    },
+    {
+      "epoch": 1.4196824655615223,
+      "grad_norm": 0.028488879438601067,
+      "learning_rate": 0.000153263943044609,
+      "loss": 1.792,
+      "step": 1520
+    },
+    {
+      "epoch": 1.429021713752043,
+      "grad_norm": 0.03125154490954804,
+      "learning_rate": 0.00015260843377746147,
+      "loss": 1.8008,
+      "step": 1530
+    },
+    {
+      "epoch": 1.4383609619425637,
+      "grad_norm": 0.030194357488801882,
+      "learning_rate": 0.00015194978386994507,
+      "loss": 1.7948,
+      "step": 1540
+    },
+    {
+      "epoch": 1.4477002101330843,
+      "grad_norm": 0.03049246845786265,
+      "learning_rate": 0.00015128803264241852,
+      "loss": 1.7967,
+      "step": 1550
+    },
+    {
+      "epoch": 1.4570394583236048,
+      "grad_norm": 0.030497211097258083,
+      "learning_rate": 0.0001506232196003844,
+      "loss": 1.7894,
+      "step": 1560
+    },
+    {
+      "epoch": 1.4663787065141256,
+      "grad_norm": 0.028748806119737205,
+      "learning_rate": 0.00014995538443213094,
+      "loss": 1.806,
+      "step": 1570
+    },
+    {
+      "epoch": 1.4757179547046464,
+      "grad_norm": 0.036423750322912396,
+      "learning_rate": 0.00014928456700636237,
+      "loss": 1.7995,
+      "step": 1580
+    },
+    {
+      "epoch": 1.485057202895167,
+      "grad_norm": 0.039101516109204065,
+      "learning_rate": 0.00014861080736981906,
+      "loss": 1.8028,
+      "step": 1590
+    },
+    {
+      "epoch": 1.4943964510856875,
+      "grad_norm": 0.031368399541673815,
+      "learning_rate": 0.00014793414574488663,
+      "loss": 1.8005,
+      "step": 1600
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 4280,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 100,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.2955222257961009e+20,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

l2-13b-ga/checkpoint-1600/zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

l2-13b-ga/checkpoint-2500/config.json

@@ -0,0 +1,30 @@
+{
+  "_name_or_path": "meta-llama/Llama-2-13b-hf",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 5120,
+  "initializer_range": 0.02,
+  "intermediate_size": 13824,
+  "max_position_embeddings": 4096,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 40,
+  "num_hidden_layers": 40,
+  "num_key_value_heads": 40,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.46.3",
+  "use_cache": true,
+  "vocab_size": 35483
+}

l2-13b-ga/checkpoint-2500/generation_config.json

@@ -0,0 +1,10 @@
+{
+  "bos_token_id": 1,
+  "do_sample": true,
+  "eos_token_id": 2,
+  "max_length": 4096,
+  "pad_token_id": 0,
+  "temperature": 0.6,
+  "top_p": 0.9,
+  "transformers_version": "4.46.3"
+}

l2-13b-ga/checkpoint-2500/latest

@@ -0,0 +1 @@
+global_step2500

l2-13b-ga/checkpoint-2500/model.safetensors.index.json

@@ -0,0 +1,370 @@
+{
+  "metadata": {
+    "total_size": 26103060480
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00006-of-00006.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.15.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.30.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.31.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.input_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.down_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.gate_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.up_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.post_attention_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.39.input_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.down_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.gate_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.up_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.post_attention_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.k_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.o_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.q_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.v_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.norm.weight": "model-00006-of-00006.safetensors"
+  }
+}

l2-13b-ga/checkpoint-2500/special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

l2-13b-ga/checkpoint-2500/tokenizer_config.json

@@ -0,0 +1,42 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": true,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": true,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

l2-13b-ga/checkpoint-2500/trainer_state.json

@@ -0,0 +1,1790 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.3350455288349288,
+  "eval_steps": 500,
+  "global_step": 2500,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0009339248190520663,
+      "grad_norm": 6.638877692627699,
+      "learning_rate": 9.345794392523364e-07,
+      "loss": 9.2917,
+      "step": 1
+    },
+    {
+      "epoch": 0.009339248190520663,
+      "grad_norm": 1.1560921335705272,
+      "learning_rate": 9.345794392523365e-06,
+      "loss": 9.0876,
+      "step": 10
+    },
+    {
+      "epoch": 0.018678496381041326,
+      "grad_norm": 0.8415132296956432,
+      "learning_rate": 1.869158878504673e-05,
+      "loss": 8.2164,
+      "step": 20
+    },
+    {
+      "epoch": 0.02801774457156199,
+      "grad_norm": 0.45381630992958155,
+      "learning_rate": 2.8037383177570094e-05,
+      "loss": 7.5184,
+      "step": 30
+    },
+    {
+      "epoch": 0.03735699276208265,
+      "grad_norm": 0.8400636107958425,
+      "learning_rate": 3.738317757009346e-05,
+      "loss": 6.6507,
+      "step": 40
+    },
+    {
+      "epoch": 0.046696240952603316,
+      "grad_norm": 0.557696240829066,
+      "learning_rate": 4.672897196261683e-05,
+      "loss": 5.8909,
+      "step": 50
+    },
+    {
+      "epoch": 0.05603548914312398,
+      "grad_norm": 0.3971996057467842,
+      "learning_rate": 5.607476635514019e-05,
+      "loss": 5.4127,
+      "step": 60
+    },
+    {
+      "epoch": 0.06537473733364464,
+      "grad_norm": 0.2932710540265688,
+      "learning_rate": 6.542056074766355e-05,
+      "loss": 5.0106,
+      "step": 70
+    },
+    {
+      "epoch": 0.0747139855241653,
+      "grad_norm": 0.3682690443551033,
+      "learning_rate": 7.476635514018692e-05,
+      "loss": 4.6042,
+      "step": 80
+    },
+    {
+      "epoch": 0.08405323371468597,
+      "grad_norm": 0.3132971920011515,
+      "learning_rate": 8.411214953271028e-05,
+      "loss": 4.2031,
+      "step": 90
+    },
+    {
+      "epoch": 0.09339248190520663,
+      "grad_norm": 0.6731868159213446,
+      "learning_rate": 9.345794392523365e-05,
+      "loss": 3.9423,
+      "step": 100
+    },
+    {
+      "epoch": 0.1027317300957273,
+      "grad_norm": 0.27848867836763197,
+      "learning_rate": 0.000102803738317757,
+      "loss": 3.7157,
+      "step": 110
+    },
+    {
+      "epoch": 0.11207097828624796,
+      "grad_norm": 0.24642109032991807,
+      "learning_rate": 0.00011214953271028037,
+      "loss": 3.4516,
+      "step": 120
+    },
+    {
+      "epoch": 0.12141022647676862,
+      "grad_norm": 0.25717384664029797,
+      "learning_rate": 0.00012149532710280373,
+      "loss": 3.2167,
+      "step": 130
+    },
+    {
+      "epoch": 0.13074947466728928,
+      "grad_norm": 0.20912922668565637,
+      "learning_rate": 0.0001308411214953271,
+      "loss": 3.0237,
+      "step": 140
+    },
+    {
+      "epoch": 0.14008872285780993,
+      "grad_norm": 0.15805888388706113,
+      "learning_rate": 0.00014018691588785047,
+      "loss": 2.8529,
+      "step": 150
+    },
+    {
+      "epoch": 0.1494279710483306,
+      "grad_norm": 0.23370349497479534,
+      "learning_rate": 0.00014953271028037384,
+      "loss": 2.7078,
+      "step": 160
+    },
+    {
+      "epoch": 0.15876721923885126,
+      "grad_norm": 0.1802138633012483,
+      "learning_rate": 0.0001588785046728972,
+      "loss": 2.6115,
+      "step": 170
+    },
+    {
+      "epoch": 0.16810646742937194,
+      "grad_norm": 0.13354347610039718,
+      "learning_rate": 0.00016822429906542056,
+      "loss": 2.5309,
+      "step": 180
+    },
+    {
+      "epoch": 0.17744571561989259,
+      "grad_norm": 0.09414865188086892,
+      "learning_rate": 0.00017757009345794393,
+      "loss": 2.4452,
+      "step": 190
+    },
+    {
+      "epoch": 0.18678496381041326,
+      "grad_norm": 0.08333601554768896,
+      "learning_rate": 0.0001869158878504673,
+      "loss": 2.3832,
+      "step": 200
+    },
+    {
+      "epoch": 0.1961242120009339,
+      "grad_norm": 0.15926414699806835,
+      "learning_rate": 0.00019626168224299065,
+      "loss": 2.3492,
+      "step": 210
+    },
+    {
+      "epoch": 0.2054634601914546,
+      "grad_norm": 0.09492820761057012,
+      "learning_rate": 0.0001999989254250208,
+      "loss": 2.323,
+      "step": 220
+    },
+    {
+      "epoch": 0.21480270838197524,
+      "grad_norm": 0.0801349259356147,
+      "learning_rate": 0.00019999235866155886,
+      "loss": 2.2731,
+      "step": 230
+    },
+    {
+      "epoch": 0.22414195657249592,
+      "grad_norm": 0.12210960524693895,
+      "learning_rate": 0.00019997982251228469,
+      "loss": 2.2433,
+      "step": 240
+    },
+    {
+      "epoch": 0.23348120476301656,
+      "grad_norm": 3.14289498732125,
+      "learning_rate": 0.00019996131772558666,
+      "loss": 3.2769,
+      "step": 250
+    },
+    {
+      "epoch": 0.24282045295353724,
+      "grad_norm": 1.632940983166179,
+      "learning_rate": 0.00019993684540617132,
+      "loss": 4.9343,
+      "step": 260
+    },
+    {
+      "epoch": 0.2521597011440579,
+      "grad_norm": 3.4831252230225416,
+      "learning_rate": 0.00019990640701499736,
+      "loss": 4.2768,
+      "step": 270
+    },
+    {
+      "epoch": 0.26149894933457857,
+      "grad_norm": 1.6069045920523788,
+      "learning_rate": 0.00019987000436918874,
+      "loss": 5.9581,
+      "step": 280
+    },
+    {
+      "epoch": 0.27083819752509924,
+      "grad_norm": 0.2220907936615993,
+      "learning_rate": 0.00019982763964192585,
+      "loss": 3.8228,
+      "step": 290
+    },
+    {
+      "epoch": 0.28017744571561987,
+      "grad_norm": 0.24737284913291765,
+      "learning_rate": 0.00019977931536231596,
+      "loss": 3.1413,
+      "step": 300
+    },
+    {
+      "epoch": 0.28951669390614054,
+      "grad_norm": 4.010404518241152,
+      "learning_rate": 0.00019972503441524224,
+      "loss": 2.8432,
+      "step": 310
+    },
+    {
+      "epoch": 0.2988559420966612,
+      "grad_norm": 0.1515583580811596,
+      "learning_rate": 0.00019966480004119142,
+      "loss": 2.7859,
+      "step": 320
+    },
+    {
+      "epoch": 0.3081951902871819,
+      "grad_norm": 0.11259395750650594,
+      "learning_rate": 0.00019959861583606045,
+      "loss": 2.5821,
+      "step": 330
+    },
+    {
+      "epoch": 0.3175344384777025,
+      "grad_norm": 0.22514797814956813,
+      "learning_rate": 0.00019952648575094183,
+      "loss": 2.4517,
+      "step": 340
+    },
+    {
+      "epoch": 0.3268736866682232,
+      "grad_norm": 0.08040136172033542,
+      "learning_rate": 0.00019944841409188767,
+      "loss": 2.3794,
+      "step": 350
+    },
+    {
+      "epoch": 0.3362129348587439,
+      "grad_norm": 0.054758073593565354,
+      "learning_rate": 0.00019936440551965263,
+      "loss": 2.3232,
+      "step": 360
+    },
+    {
+      "epoch": 0.34555218304926455,
+      "grad_norm": 0.06742998909645591,
+      "learning_rate": 0.00019927446504941577,
+      "loss": 2.2776,
+      "step": 370
+    },
+    {
+      "epoch": 0.35489143123978517,
+      "grad_norm": 0.048780907584876736,
+      "learning_rate": 0.00019917859805048096,
+      "loss": 2.2376,
+      "step": 380
+    },
+    {
+      "epoch": 0.36423067943030585,
+      "grad_norm": 0.0475325963052214,
+      "learning_rate": 0.00019907681024595663,
+      "loss": 2.2191,
+      "step": 390
+    },
+    {
+      "epoch": 0.3735699276208265,
+      "grad_norm": 0.054089563211590065,
+      "learning_rate": 0.00019896910771241387,
+      "loss": 2.1961,
+      "step": 400
+    },
+    {
+      "epoch": 0.3829091758113472,
+      "grad_norm": 0.21798406131864823,
+      "learning_rate": 0.00019885549687952372,
+      "loss": 2.2078,
+      "step": 410
+    },
+    {
+      "epoch": 0.3922484240018678,
+      "grad_norm": 0.8673185709111124,
+      "learning_rate": 0.00019873598452967338,
+      "loss": 2.3731,
+      "step": 420
+    },
+    {
+      "epoch": 0.4015876721923885,
+      "grad_norm": 0.22424350669971718,
+      "learning_rate": 0.0001986105777975613,
+      "loss": 2.6195,
+      "step": 430
+    },
+    {
+      "epoch": 0.4109269203829092,
+      "grad_norm": 0.307418135168262,
+      "learning_rate": 0.00019847928416977126,
+      "loss": 2.3624,
+      "step": 440
+    },
+    {
+      "epoch": 0.42026616857342985,
+      "grad_norm": 0.07944722668080402,
+      "learning_rate": 0.00019834211148432536,
+      "loss": 2.2799,
+      "step": 450
+    },
+    {
+      "epoch": 0.4296054167639505,
+      "grad_norm": 0.18146933758664588,
+      "learning_rate": 0.00019819906793021614,
+      "loss": 2.2177,
+      "step": 460
+    },
+    {
+      "epoch": 0.43894466495447115,
+      "grad_norm": 0.07035825837333018,
+      "learning_rate": 0.0001980501620469178,
+      "loss": 2.1767,
+      "step": 470
+    },
+    {
+      "epoch": 0.44828391314499183,
+      "grad_norm": 0.04596186944454228,
+      "learning_rate": 0.0001978954027238763,
+      "loss": 2.1598,
+      "step": 480
+    },
+    {
+      "epoch": 0.4576231613355125,
+      "grad_norm": 0.041342347745088055,
+      "learning_rate": 0.0001977347991999786,
+      "loss": 2.131,
+      "step": 490
+    },
+    {
+      "epoch": 0.46696240952603313,
+      "grad_norm": 0.04172063219841485,
+      "learning_rate": 0.00019756836106300137,
+      "loss": 2.1231,
+      "step": 500
+    },
+    {
+      "epoch": 0.4763016577165538,
+      "grad_norm": 0.03373646457711144,
+      "learning_rate": 0.00019739609824903843,
+      "loss": 2.1146,
+      "step": 510
+    },
+    {
+      "epoch": 0.4856409059070745,
+      "grad_norm": 0.03736871030676605,
+      "learning_rate": 0.00019721802104190748,
+      "loss": 2.1003,
+      "step": 520
+    },
+    {
+      "epoch": 0.49498015409759516,
+      "grad_norm": 0.033931028038211034,
+      "learning_rate": 0.00019703414007253645,
+      "loss": 2.0983,
+      "step": 530
+    },
+    {
+      "epoch": 0.5043194022881158,
+      "grad_norm": 0.03790055446070549,
+      "learning_rate": 0.00019684446631832868,
+      "loss": 2.092,
+      "step": 540
+    },
+    {
+      "epoch": 0.5136586504786365,
+      "grad_norm": 0.030956192803893078,
+      "learning_rate": 0.00019664901110250758,
+      "loss": 2.0807,
+      "step": 550
+    },
+    {
+      "epoch": 0.5229978986691571,
+      "grad_norm": 0.03542530209935129,
+      "learning_rate": 0.00019644778609344068,
+      "loss": 2.0773,
+      "step": 560
+    },
+    {
+      "epoch": 0.5323371468596778,
+      "grad_norm": 0.040947757568902336,
+      "learning_rate": 0.00019624080330394306,
+      "loss": 2.0649,
+      "step": 570
+    },
+    {
+      "epoch": 0.5416763950501985,
+      "grad_norm": 0.034273415973688146,
+      "learning_rate": 0.00019602807509056018,
+      "loss": 2.0479,
+      "step": 580
+    },
+    {
+      "epoch": 0.5510156432407192,
+      "grad_norm": 0.031427481498873144,
+      "learning_rate": 0.00019580961415283028,
+      "loss": 2.0563,
+      "step": 590
+    },
+    {
+      "epoch": 0.5603548914312397,
+      "grad_norm": 0.03141549752041532,
+      "learning_rate": 0.00019558543353252611,
+      "loss": 2.0503,
+      "step": 600
+    },
+    {
+      "epoch": 0.5696941396217604,
+      "grad_norm": 0.033012392726428204,
+      "learning_rate": 0.00019535554661287652,
+      "loss": 2.0389,
+      "step": 610
+    },
+    {
+      "epoch": 0.5790333878122811,
+      "grad_norm": 0.02913261992661444,
+      "learning_rate": 0.0001951199671177673,
+      "loss": 2.036,
+      "step": 620
+    },
+    {
+      "epoch": 0.5883726360028018,
+      "grad_norm": 0.030543903708435332,
+      "learning_rate": 0.00019487870911092214,
+      "loss": 2.0326,
+      "step": 630
+    },
+    {
+      "epoch": 0.5977118841933224,
+      "grad_norm": 0.03215005545393897,
+      "learning_rate": 0.00019463178699506277,
+      "loss": 2.0231,
+      "step": 640
+    },
+    {
+      "epoch": 0.6070511323838431,
+      "grad_norm": 0.03823630791937631,
+      "learning_rate": 0.00019437921551104933,
+      "loss": 2.0293,
+      "step": 650
+    },
+    {
+      "epoch": 0.6163903805743638,
+      "grad_norm": 0.03200103149471209,
+      "learning_rate": 0.00019412100973700038,
+      "loss": 2.017,
+      "step": 660
+    },
+    {
+      "epoch": 0.6257296287648845,
+      "grad_norm": 0.03841804153577787,
+      "learning_rate": 0.00019385718508739262,
+      "loss": 2.0135,
+      "step": 670
+    },
+    {
+      "epoch": 0.635068876955405,
+      "grad_norm": 0.03052396655271533,
+      "learning_rate": 0.0001935877573121407,
+      "loss": 2.0237,
+      "step": 680
+    },
+    {
+      "epoch": 0.6444081251459257,
+      "grad_norm": 0.033817837533771815,
+      "learning_rate": 0.00019331274249565717,
+      "loss": 2.0069,
+      "step": 690
+    },
+    {
+      "epoch": 0.6537473733364464,
+      "grad_norm": 0.028286999650643876,
+      "learning_rate": 0.00019303215705589194,
+      "loss": 2.0112,
+      "step": 700
+    },
+    {
+      "epoch": 0.6630866215269671,
+      "grad_norm": 0.030693948626257357,
+      "learning_rate": 0.00019274601774335243,
+      "loss": 2.0,
+      "step": 710
+    },
+    {
+      "epoch": 0.6724258697174877,
+      "grad_norm": 0.029984615635953022,
+      "learning_rate": 0.0001924543416401035,
+      "loss": 2.0028,
+      "step": 720
+    },
+    {
+      "epoch": 0.6817651179080084,
+      "grad_norm": 0.02915985673921391,
+      "learning_rate": 0.00019215714615874755,
+      "loss": 2.0031,
+      "step": 730
+    },
+    {
+      "epoch": 0.6911043660985291,
+      "grad_norm": 0.028305399777245336,
+      "learning_rate": 0.00019185444904138528,
+      "loss": 1.9924,
+      "step": 740
+    },
+    {
+      "epoch": 0.7004436142890498,
+      "grad_norm": 0.036720505429756495,
+      "learning_rate": 0.00019154626835855628,
+      "loss": 1.9981,
+      "step": 750
+    },
+    {
+      "epoch": 0.7097828624795703,
+      "grad_norm": 0.0287124048917296,
+      "learning_rate": 0.00019123262250816034,
+      "loss": 1.9868,
+      "step": 760
+    },
+    {
+      "epoch": 0.719122110670091,
+      "grad_norm": 0.03318092492837997,
+      "learning_rate": 0.00019091353021435915,
+      "loss": 1.9943,
+      "step": 770
+    },
+    {
+      "epoch": 0.7284613588606117,
+      "grad_norm": 0.054035272137015325,
+      "learning_rate": 0.00019058901052645844,
+      "loss": 1.9838,
+      "step": 780
+    },
+    {
+      "epoch": 0.7378006070511324,
+      "grad_norm": 0.03184392761983255,
+      "learning_rate": 0.00019025908281777078,
+      "loss": 1.982,
+      "step": 790
+    },
+    {
+      "epoch": 0.747139855241653,
+      "grad_norm": 0.029409948164434735,
+      "learning_rate": 0.00018992376678445908,
+      "loss": 1.9693,
+      "step": 800
+    },
+    {
+      "epoch": 0.7564791034321737,
+      "grad_norm": 0.029656963043919016,
+      "learning_rate": 0.00018958308244436064,
+      "loss": 1.9914,
+      "step": 810
+    },
+    {
+      "epoch": 0.7658183516226944,
+      "grad_norm": 0.030843610865326686,
+      "learning_rate": 0.00018923705013579233,
+      "loss": 1.9749,
+      "step": 820
+    },
+    {
+      "epoch": 0.7751575998132151,
+      "grad_norm": 0.0377584286045999,
+      "learning_rate": 0.00018888569051633613,
+      "loss": 1.9606,
+      "step": 830
+    },
+    {
+      "epoch": 0.7844968480037356,
+      "grad_norm": 0.03147699916274391,
+      "learning_rate": 0.00018852902456160616,
+      "loss": 1.9696,
+      "step": 840
+    },
+    {
+      "epoch": 0.7938360961942563,
+      "grad_norm": 0.029193153251471263,
+      "learning_rate": 0.0001881670735639963,
+      "loss": 1.9687,
+      "step": 850
+    },
+    {
+      "epoch": 0.803175344384777,
+      "grad_norm": 0.030201895228156087,
+      "learning_rate": 0.00018779985913140924,
+      "loss": 1.9678,
+      "step": 860
+    },
+    {
+      "epoch": 0.8125145925752977,
+      "grad_norm": 0.029272319485493213,
+      "learning_rate": 0.00018742740318596632,
+      "loss": 1.9697,
+      "step": 870
+    },
+    {
+      "epoch": 0.8218538407658184,
+      "grad_norm": 0.033740176465285654,
+      "learning_rate": 0.000187049727962699,
+      "loss": 1.9647,
+      "step": 880
+    },
+    {
+      "epoch": 0.831193088956339,
+      "grad_norm": 0.029539399251208593,
+      "learning_rate": 0.0001866668560082213,
+      "loss": 1.9627,
+      "step": 890
+    },
+    {
+      "epoch": 0.8405323371468597,
+      "grad_norm": 0.028666077337951026,
+      "learning_rate": 0.0001862788101793839,
+      "loss": 1.9529,
+      "step": 900
+    },
+    {
+      "epoch": 0.8498715853373804,
+      "grad_norm": 0.030280792492665805,
+      "learning_rate": 0.0001858856136419097,
+      "loss": 1.9536,
+      "step": 910
+    },
+    {
+      "epoch": 0.859210833527901,
+      "grad_norm": 0.03921536105057096,
+      "learning_rate": 0.0001854872898690106,
+      "loss": 1.9474,
+      "step": 920
+    },
+    {
+      "epoch": 0.8685500817184216,
+      "grad_norm": 0.030632523637038354,
+      "learning_rate": 0.0001850838626399865,
+      "loss": 1.9423,
+      "step": 930
+    },
+    {
+      "epoch": 0.8778893299089423,
+      "grad_norm": 0.04615147601979514,
+      "learning_rate": 0.00018467535603880548,
+      "loss": 1.946,
+      "step": 940
+    },
+    {
+      "epoch": 0.887228578099463,
+      "grad_norm": 0.028216236017006333,
+      "learning_rate": 0.00018426179445266616,
+      "loss": 1.9408,
+      "step": 950
+    },
+    {
+      "epoch": 0.8965678262899837,
+      "grad_norm": 0.0282407563402959,
+      "learning_rate": 0.00018384320257054177,
+      "loss": 1.9447,
+      "step": 960
+    },
+    {
+      "epoch": 0.9059070744805043,
+      "grad_norm": 0.029365880854687894,
+      "learning_rate": 0.0001834196053817062,
+      "loss": 1.9389,
+      "step": 970
+    },
+    {
+      "epoch": 0.915246322671025,
+      "grad_norm": 0.02855404439130719,
+      "learning_rate": 0.00018299102817424234,
+      "loss": 1.9425,
+      "step": 980
+    },
+    {
+      "epoch": 0.9245855708615457,
+      "grad_norm": 0.03238310984070135,
+      "learning_rate": 0.00018255749653353225,
+      "loss": 1.9392,
+      "step": 990
+    },
+    {
+      "epoch": 0.9339248190520663,
+      "grad_norm": 0.030115350805430388,
+      "learning_rate": 0.00018211903634072983,
+      "loss": 1.956,
+      "step": 1000
+    },
+    {
+      "epoch": 0.9432640672425869,
+      "grad_norm": 0.039090564032501135,
+      "learning_rate": 0.0001816756737712158,
+      "loss": 1.9358,
+      "step": 1010
+    },
+    {
+      "epoch": 0.9526033154331076,
+      "grad_norm": 0.028567954476327023,
+      "learning_rate": 0.000181227435293035,
+      "loss": 1.9342,
+      "step": 1020
+    },
+    {
+      "epoch": 0.9619425636236283,
+      "grad_norm": 0.028158653241284505,
+      "learning_rate": 0.00018077434766531624,
+      "loss": 1.9287,
+      "step": 1030
+    },
+    {
+      "epoch": 0.971281811814149,
+      "grad_norm": 0.02927795717651538,
+      "learning_rate": 0.00018031643793667504,
+      "loss": 1.9298,
+      "step": 1040
+    },
+    {
+      "epoch": 0.9806210600046696,
+      "grad_norm": 0.06298340584032344,
+      "learning_rate": 0.0001798537334435986,
+      "loss": 1.9303,
+      "step": 1050
+    },
+    {
+      "epoch": 0.9899603081951903,
+      "grad_norm": 0.03627278406983255,
+      "learning_rate": 0.00017938626180881407,
+      "loss": 1.9285,
+      "step": 1060
+    },
+    {
+      "epoch": 0.999299556385711,
+      "grad_norm": 0.03916823421329747,
+      "learning_rate": 0.00017891405093963938,
+      "loss": 1.9239,
+      "step": 1070
+    },
+    {
+      "epoch": 1.0087555451786132,
+      "grad_norm": 0.03383539251115568,
+      "learning_rate": 0.00017843712902631723,
+      "loss": 1.8855,
+      "step": 1080
+    },
+    {
+      "epoch": 1.0180947933691338,
+      "grad_norm": 0.02931021936560147,
+      "learning_rate": 0.00017795552454033224,
+      "loss": 1.8004,
+      "step": 1090
+    },
+    {
+      "epoch": 1.0274340415596543,
+      "grad_norm": 0.029173295095350292,
+      "learning_rate": 0.0001774692662327113,
+      "loss": 1.7912,
+      "step": 1100
+    },
+    {
+      "epoch": 1.0367732897501751,
+      "grad_norm": 0.035210065642974735,
+      "learning_rate": 0.000176978383132307,
+      "loss": 1.7902,
+      "step": 1110
+    },
+    {
+      "epoch": 1.0461125379406957,
+      "grad_norm": 0.029794447659573477,
+      "learning_rate": 0.00017648290454406475,
+      "loss": 1.8072,
+      "step": 1120
+    },
+    {
+      "epoch": 1.0554517861312165,
+      "grad_norm": 0.03194584938279939,
+      "learning_rate": 0.0001759828600472734,
+      "loss": 1.803,
+      "step": 1130
+    },
+    {
+      "epoch": 1.064791034321737,
+      "grad_norm": 0.031028415312581603,
+      "learning_rate": 0.00017547827949379924,
+      "loss": 1.7945,
+      "step": 1140
+    },
+    {
+      "epoch": 1.0741302825122578,
+      "grad_norm": 0.03313245828751237,
+      "learning_rate": 0.00017496919300630403,
+      "loss": 1.8139,
+      "step": 1150
+    },
+    {
+      "epoch": 1.0834695307027784,
+      "grad_norm": 0.03049431724979126,
+      "learning_rate": 0.00017445563097644664,
+      "loss": 1.8031,
+      "step": 1160
+    },
+    {
+      "epoch": 1.0928087788932992,
+      "grad_norm": 0.02880548705343715,
+      "learning_rate": 0.00017393762406306878,
+      "loss": 1.7974,
+      "step": 1170
+    },
+    {
+      "epoch": 1.1021480270838198,
+      "grad_norm": 0.03200427895977668,
+      "learning_rate": 0.00017341520319036469,
+      "loss": 1.7994,
+      "step": 1180
+    },
+    {
+      "epoch": 1.1114872752743403,
+      "grad_norm": 0.031293532815600045,
+      "learning_rate": 0.00017288839954603496,
+      "loss": 1.8073,
+      "step": 1190
+    },
+    {
+      "epoch": 1.1208265234648611,
+      "grad_norm": 0.032367211109345505,
+      "learning_rate": 0.00017235724457942468,
+      "loss": 1.7944,
+      "step": 1200
+    },
+    {
+      "epoch": 1.1301657716553817,
+      "grad_norm": 0.037783793923191374,
+      "learning_rate": 0.0001718217699996462,
+      "loss": 1.7948,
+      "step": 1210
+    },
+    {
+      "epoch": 1.1395050198459025,
+      "grad_norm": 0.02881083117349317,
+      "learning_rate": 0.00017128200777368567,
+      "loss": 1.8029,
+      "step": 1220
+    },
+    {
+      "epoch": 1.148844268036423,
+      "grad_norm": 0.032997408408175985,
+      "learning_rate": 0.00017073799012449524,
+      "loss": 1.7914,
+      "step": 1230
+    },
+    {
+      "epoch": 1.1581835162269438,
+      "grad_norm": 0.030263615801190885,
+      "learning_rate": 0.00017018974952906884,
+      "loss": 1.792,
+      "step": 1240
+    },
+    {
+      "epoch": 1.1675227644174644,
+      "grad_norm": 0.03286310429098484,
+      "learning_rate": 0.00016963731871650378,
+      "loss": 1.8149,
+      "step": 1250
+    },
+    {
+      "epoch": 1.1768620126079852,
+      "grad_norm": 0.031111621587597126,
+      "learning_rate": 0.00016908073066604663,
+      "loss": 1.8092,
+      "step": 1260
+    },
+    {
+      "epoch": 1.1862012607985057,
+      "grad_norm": 0.03372950172075241,
+      "learning_rate": 0.0001685200186051246,
+      "loss": 1.818,
+      "step": 1270
+    },
+    {
+      "epoch": 1.1955405089890263,
+      "grad_norm": 0.040782040436567434,
+      "learning_rate": 0.00016795521600736164,
+      "loss": 1.7999,
+      "step": 1280
+    },
+    {
+      "epoch": 1.204879757179547,
+      "grad_norm": 0.03451605168178924,
+      "learning_rate": 0.00016738635659058044,
+      "loss": 1.7945,
+      "step": 1290
+    },
+    {
+      "epoch": 1.2142190053700677,
+      "grad_norm": 0.03235681588882673,
+      "learning_rate": 0.00016681347431478933,
+      "loss": 1.8087,
+      "step": 1300
+    },
+    {
+      "epoch": 1.2235582535605884,
+      "grad_norm": 0.030750745605971932,
+      "learning_rate": 0.00016623660338015487,
+      "loss": 1.7995,
+      "step": 1310
+    },
+    {
+      "epoch": 1.232897501751109,
+      "grad_norm": 0.029444668665577274,
+      "learning_rate": 0.00016565577822496042,
+      "loss": 1.8025,
+      "step": 1320
+    },
+    {
+      "epoch": 1.2422367499416298,
+      "grad_norm": 0.038528856709584745,
+      "learning_rate": 0.00016507103352354996,
+      "loss": 1.7954,
+      "step": 1330
+    },
+    {
+      "epoch": 1.2515759981321504,
+      "grad_norm": 0.034217088004383035,
+      "learning_rate": 0.00016448240418425814,
+      "loss": 1.7962,
+      "step": 1340
+    },
+    {
+      "epoch": 1.260915246322671,
+      "grad_norm": 0.030205405393195585,
+      "learning_rate": 0.00016388992534732645,
+      "loss": 1.7973,
+      "step": 1350
+    },
+    {
+      "epoch": 1.2702544945131917,
+      "grad_norm": 0.029082218516562994,
+      "learning_rate": 0.00016329363238280528,
+      "loss": 1.796,
+      "step": 1360
+    },
+    {
+      "epoch": 1.2795937427037123,
+      "grad_norm": 0.029003887688766505,
+      "learning_rate": 0.00016269356088844238,
+      "loss": 1.7946,
+      "step": 1370
+    },
+    {
+      "epoch": 1.288932990894233,
+      "grad_norm": 0.03341157363649238,
+      "learning_rate": 0.00016208974668755779,
+      "loss": 1.7972,
+      "step": 1380
+    },
+    {
+      "epoch": 1.2982722390847536,
+      "grad_norm": 0.030614480844663026,
+      "learning_rate": 0.00016148222582690517,
+      "loss": 1.7973,
+      "step": 1390
+    },
+    {
+      "epoch": 1.3076114872752744,
+      "grad_norm": 0.029741346740467405,
+      "learning_rate": 0.00016087103457452,
+      "loss": 1.8076,
+      "step": 1400
+    },
+    {
+      "epoch": 1.316950735465795,
+      "grad_norm": 0.029569313554185597,
+      "learning_rate": 0.00016025620941755424,
+      "loss": 1.8043,
+      "step": 1410
+    },
+    {
+      "epoch": 1.3262899836563156,
+      "grad_norm": 0.02947637404374054,
+      "learning_rate": 0.0001596377870600983,
+      "loss": 1.797,
+      "step": 1420
+    },
+    {
+      "epoch": 1.3356292318468364,
+      "grad_norm": 0.031005062093959545,
+      "learning_rate": 0.00015901580442098968,
+      "loss": 1.8086,
+      "step": 1430
+    },
+    {
+      "epoch": 1.344968480037357,
+      "grad_norm": 0.029493792984873927,
+      "learning_rate": 0.00015839029863160922,
+      "loss": 1.8026,
+      "step": 1440
+    },
+    {
+      "epoch": 1.3543077282278777,
+      "grad_norm": 0.0288068155951218,
+      "learning_rate": 0.0001577613070336641,
+      "loss": 1.7951,
+      "step": 1450
+    },
+    {
+      "epoch": 1.3636469764183983,
+      "grad_norm": 0.03380404824627639,
+      "learning_rate": 0.00015712886717695885,
+      "loss": 1.7938,
+      "step": 1460
+    },
+    {
+      "epoch": 1.372986224608919,
+      "grad_norm": 0.039744706189693335,
+      "learning_rate": 0.0001564930168171536,
+      "loss": 1.8016,
+      "step": 1470
+    },
+    {
+      "epoch": 1.3823254727994396,
+      "grad_norm": 0.030565530594285437,
+      "learning_rate": 0.00015585379391351012,
+      "loss": 1.7984,
+      "step": 1480
+    },
+    {
+      "epoch": 1.3916647209899602,
+      "grad_norm": 0.04009392805554255,
+      "learning_rate": 0.00015521123662662567,
+      "loss": 1.7999,
+      "step": 1490
+    },
+    {
+      "epoch": 1.401003969180481,
+      "grad_norm": 0.03516196009586836,
+      "learning_rate": 0.000154565383316155,
+      "loss": 1.7979,
+      "step": 1500
+    },
+    {
+      "epoch": 1.4103432173710018,
+      "grad_norm": 0.03534161399054556,
+      "learning_rate": 0.0001539162725385202,
+      "loss": 1.8057,
+      "step": 1510
+    },
+    {
+      "epoch": 1.4196824655615223,
+      "grad_norm": 0.028488879438601067,
+      "learning_rate": 0.000153263943044609,
+      "loss": 1.792,
+      "step": 1520
+    },
+    {
+      "epoch": 1.429021713752043,
+      "grad_norm": 0.03125154490954804,
+      "learning_rate": 0.00015260843377746147,
+      "loss": 1.8008,
+      "step": 1530
+    },
+    {
+      "epoch": 1.4383609619425637,
+      "grad_norm": 0.030194357488801882,
+      "learning_rate": 0.00015194978386994507,
+      "loss": 1.7948,
+      "step": 1540
+    },
+    {
+      "epoch": 1.4477002101330843,
+      "grad_norm": 0.03049246845786265,
+      "learning_rate": 0.00015128803264241852,
+      "loss": 1.7967,
+      "step": 1550
+    },
+    {
+      "epoch": 1.4570394583236048,
+      "grad_norm": 0.030497211097258083,
+      "learning_rate": 0.0001506232196003844,
+      "loss": 1.7894,
+      "step": 1560
+    },
+    {
+      "epoch": 1.4663787065141256,
+      "grad_norm": 0.028748806119737205,
+      "learning_rate": 0.00014995538443213094,
+      "loss": 1.806,
+      "step": 1570
+    },
+    {
+      "epoch": 1.4757179547046464,
+      "grad_norm": 0.036423750322912396,
+      "learning_rate": 0.00014928456700636237,
+      "loss": 1.7995,
+      "step": 1580
+    },
+    {
+      "epoch": 1.485057202895167,
+      "grad_norm": 0.039101516109204065,
+      "learning_rate": 0.00014861080736981906,
+      "loss": 1.8028,
+      "step": 1590
+    },
+    {
+      "epoch": 1.4943964510856875,
+      "grad_norm": 0.031368399541673815,
+      "learning_rate": 0.00014793414574488663,
+      "loss": 1.8005,
+      "step": 1600
+    },
+    {
+      "epoch": 1.5037356992762083,
+      "grad_norm": 0.029788484702512056,
+      "learning_rate": 0.00014725462252719495,
+      "loss": 1.7963,
+      "step": 1610
+    },
+    {
+      "epoch": 1.5130749474667289,
+      "grad_norm": 0.029719041811636312,
+      "learning_rate": 0.00014657227828320635,
+      "loss": 1.7957,
+      "step": 1620
+    },
+    {
+      "epoch": 1.5224141956572494,
+      "grad_norm": 0.02820041575417432,
+      "learning_rate": 0.00014588715374779407,
+      "loss": 1.7986,
+      "step": 1630
+    },
+    {
+      "epoch": 1.5317534438477702,
+      "grad_norm": 1.8322544897261024,
+      "learning_rate": 0.0001451992898218102,
+      "loss": 2.016,
+      "step": 1640
+    },
+    {
+      "epoch": 1.541092692038291,
+      "grad_norm": 0.8676995793107466,
+      "learning_rate": 0.0001445087275696443,
+      "loss": 2.1095,
+      "step": 1650
+    },
+    {
+      "epoch": 1.5504319402288116,
+      "grad_norm": 0.08688193434631736,
+      "learning_rate": 0.00014381550821677155,
+      "loss": 2.0497,
+      "step": 1660
+    },
+    {
+      "epoch": 1.5597711884193322,
+      "grad_norm": 0.18321381224589608,
+      "learning_rate": 0.0001431196731472921,
+      "loss": 1.9023,
+      "step": 1670
+    },
+    {
+      "epoch": 1.569110436609853,
+      "grad_norm": 0.058214343698110564,
+      "learning_rate": 0.00014242126390145998,
+      "loss": 1.8566,
+      "step": 1680
+    },
+    {
+      "epoch": 1.5784496848003737,
+      "grad_norm": 0.03965578147557666,
+      "learning_rate": 0.0001417203221732036,
+      "loss": 1.8206,
+      "step": 1690
+    },
+    {
+      "epoch": 1.587788932990894,
+      "grad_norm": 0.03131802880017099,
+      "learning_rate": 0.00014101688980763658,
+      "loss": 1.8272,
+      "step": 1700
+    },
+    {
+      "epoch": 1.5971281811814149,
+      "grad_norm": 0.02788722706683908,
+      "learning_rate": 0.00014031100879855968,
+      "loss": 1.8145,
+      "step": 1710
+    },
+    {
+      "epoch": 1.6064674293719357,
+      "grad_norm": 0.02781057130092059,
+      "learning_rate": 0.00013960272128595372,
+      "loss": 1.8122,
+      "step": 1720
+    },
+    {
+      "epoch": 1.6158066775624562,
+      "grad_norm": 0.032067383734154756,
+      "learning_rate": 0.00013889206955346403,
+      "loss": 1.8064,
+      "step": 1730
+    },
+    {
+      "epoch": 1.6251459257529768,
+      "grad_norm": 0.03448807133884858,
+      "learning_rate": 0.00013817909602587613,
+      "loss": 1.8128,
+      "step": 1740
+    },
+    {
+      "epoch": 1.6344851739434976,
+      "grad_norm": 0.04127395288271696,
+      "learning_rate": 0.00013746384326658305,
+      "loss": 1.8041,
+      "step": 1750
+    },
+    {
+      "epoch": 1.6438244221340184,
+      "grad_norm": 0.030243192787820814,
+      "learning_rate": 0.00013674635397504427,
+      "loss": 1.803,
+      "step": 1760
+    },
+    {
+      "epoch": 1.653163670324539,
+      "grad_norm": 0.032183525625428915,
+      "learning_rate": 0.00013602667098423687,
+      "loss": 1.8092,
+      "step": 1770
+    },
+    {
+      "epoch": 1.6625029185150595,
+      "grad_norm": 0.030391044676815377,
+      "learning_rate": 0.00013530483725809818,
+      "loss": 1.8039,
+      "step": 1780
+    },
+    {
+      "epoch": 1.6718421667055803,
+      "grad_norm": 0.03204756818238517,
+      "learning_rate": 0.000134580895888961,
+      "loss": 1.8017,
+      "step": 1790
+    },
+    {
+      "epoch": 1.6811814148961008,
+      "grad_norm": 0.031117112662107062,
+      "learning_rate": 0.00013385489009498124,
+      "loss": 1.8017,
+      "step": 1800
+    },
+    {
+      "epoch": 1.6905206630866214,
+      "grad_norm": 0.028389293445401805,
+      "learning_rate": 0.00013312686321755761,
+      "loss": 1.811,
+      "step": 1810
+    },
+    {
+      "epoch": 1.6998599112771422,
+      "grad_norm": 4.908185097372493,
+      "learning_rate": 0.0001323968587187443,
+      "loss": 2.194,
+      "step": 1820
+    },
+    {
+      "epoch": 1.709199159467663,
+      "grad_norm": 0.2849452041194025,
+      "learning_rate": 0.00013166492017865637,
+      "loss": 2.0785,
+      "step": 1830
+    },
+    {
+      "epoch": 1.7185384076581836,
+      "grad_norm": 0.10989252058989733,
+      "learning_rate": 0.0001309310912928682,
+      "loss": 1.986,
+      "step": 1840
+    },
+    {
+      "epoch": 1.7278776558487041,
+      "grad_norm": 0.057475612656740484,
+      "learning_rate": 0.00013019541586980463,
+      "loss": 1.8614,
+      "step": 1850
+    },
+    {
+      "epoch": 1.737216904039225,
+      "grad_norm": 0.034908181734207726,
+      "learning_rate": 0.000129457937828126,
+      "loss": 1.8326,
+      "step": 1860
+    },
+    {
+      "epoch": 1.7465561522297455,
+      "grad_norm": 0.02892836681897248,
+      "learning_rate": 0.00012871870119410614,
+      "loss": 1.8243,
+      "step": 1870
+    },
+    {
+      "epoch": 1.755895400420266,
+      "grad_norm": 0.03311636981729384,
+      "learning_rate": 0.00012797775009900397,
+      "loss": 1.8183,
+      "step": 1880
+    },
+    {
+      "epoch": 1.7652346486107868,
+      "grad_norm": 0.03151917952358458,
+      "learning_rate": 0.00012723512877642904,
+      "loss": 1.8034,
+      "step": 1890
+    },
+    {
+      "epoch": 1.7745738968013076,
+      "grad_norm": 0.028109921832296925,
+      "learning_rate": 0.000126490881559701,
+      "loss": 1.8129,
+      "step": 1900
+    },
+    {
+      "epoch": 1.7839131449918282,
+      "grad_norm": 0.030350462454962698,
+      "learning_rate": 0.00012574505287920259,
+      "loss": 1.8003,
+      "step": 1910
+    },
+    {
+      "epoch": 1.7932523931823487,
+      "grad_norm": 0.03131380630103849,
+      "learning_rate": 0.00012499768725972754,
+      "loss": 1.814,
+      "step": 1920
+    },
+    {
+      "epoch": 1.8025916413728695,
+      "grad_norm": 0.029450198273050322,
+      "learning_rate": 0.00012424882931782243,
+      "loss": 1.7998,
+      "step": 1930
+    },
+    {
+      "epoch": 1.81193088956339,
+      "grad_norm": 0.0310524261453681,
+      "learning_rate": 0.0001234985237591231,
+      "loss": 1.8078,
+      "step": 1940
+    },
+    {
+      "epoch": 1.8212701377539107,
+      "grad_norm": 0.029362038478982822,
+      "learning_rate": 0.00012274681537568585,
+      "loss": 1.8014,
+      "step": 1950
+    },
+    {
+      "epoch": 1.8306093859444315,
+      "grad_norm": 0.027166816460226118,
+      "learning_rate": 0.00012199374904331337,
+      "loss": 1.8021,
+      "step": 1960
+    },
+    {
+      "epoch": 1.8399486341349522,
+      "grad_norm": 0.03680509283276228,
+      "learning_rate": 0.00012123936971887578,
+      "loss": 1.7973,
+      "step": 1970
+    },
+    {
+      "epoch": 1.8492878823254728,
+      "grad_norm": 0.03135174840346185,
+      "learning_rate": 0.0001204837224376267,
+      "loss": 1.7874,
+      "step": 1980
+    },
+    {
+      "epoch": 1.8586271305159934,
+      "grad_norm": 0.02644533256389969,
+      "learning_rate": 0.0001197268523105148,
+      "loss": 1.798,
+      "step": 1990
+    },
+    {
+      "epoch": 1.8679663787065142,
+      "grad_norm": 0.02999453651649614,
+      "learning_rate": 0.00011896880452149077,
+      "loss": 1.7957,
+      "step": 2000
+    },
+    {
+      "epoch": 1.877305626897035,
+      "grad_norm": 0.026905209700322272,
+      "learning_rate": 0.00011820962432480985,
+      "loss": 1.793,
+      "step": 2010
+    },
+    {
+      "epoch": 1.8866448750875553,
+      "grad_norm": 0.027263640323285022,
+      "learning_rate": 0.00011744935704233005,
+      "loss": 1.7974,
+      "step": 2020
+    },
+    {
+      "epoch": 1.895984123278076,
+      "grad_norm": 0.030479226063932337,
+      "learning_rate": 0.00011668804806080693,
+      "loss": 1.7898,
+      "step": 2030
+    },
+    {
+      "epoch": 1.9053233714685969,
+      "grad_norm": 0.030129902025534238,
+      "learning_rate": 0.00011592574282918369,
+      "loss": 1.7856,
+      "step": 2040
+    },
+    {
+      "epoch": 1.9146626196591174,
+      "grad_norm": 0.027884976674153635,
+      "learning_rate": 0.00011516248685587814,
+      "loss": 1.7858,
+      "step": 2050
+    },
+    {
+      "epoch": 1.924001867849638,
+      "grad_norm": 0.02925266011156687,
+      "learning_rate": 0.00011439832570606586,
+      "loss": 1.7876,
+      "step": 2060
+    },
+    {
+      "epoch": 1.9333411160401588,
+      "grad_norm": 0.028472914828616754,
+      "learning_rate": 0.00011363330499895997,
+      "loss": 1.7834,
+      "step": 2070
+    },
+    {
+      "epoch": 1.9426803642306796,
+      "grad_norm": 0.025877740032137875,
+      "learning_rate": 0.00011286747040508789,
+      "loss": 1.7955,
+      "step": 2080
+    },
+    {
+      "epoch": 1.9520196124212001,
+      "grad_norm": 0.02605295620697312,
+      "learning_rate": 0.0001121008676435648,
+      "loss": 1.7877,
+      "step": 2090
+    },
+    {
+      "epoch": 1.9613588606117207,
+      "grad_norm": 0.026887649929567867,
+      "learning_rate": 0.00011133354247936423,
+      "loss": 1.773,
+      "step": 2100
+    },
+    {
+      "epoch": 1.9706981088022415,
+      "grad_norm": 0.027982045915154026,
+      "learning_rate": 0.00011056554072058596,
+      "loss": 1.7762,
+      "step": 2110
+    },
+    {
+      "epoch": 1.980037356992762,
+      "grad_norm": 0.028077027059053006,
+      "learning_rate": 0.0001097969082157215,
+      "loss": 1.7963,
+      "step": 2120
+    },
+    {
+      "epoch": 1.9893766051832826,
+      "grad_norm": 0.027877078975954036,
+      "learning_rate": 0.00010902769085091686,
+      "loss": 1.7787,
+      "step": 2130
+    },
+    {
+      "epoch": 1.9987158533738034,
+      "grad_norm": 0.026120077388738373,
+      "learning_rate": 0.00010825793454723325,
+      "loss": 1.7842,
+      "step": 2140
+    },
+    {
+      "epoch": 2.0081718421667056,
+      "grad_norm": 0.044810283079268924,
+      "learning_rate": 0.00010748768525790569,
+      "loss": 1.6591,
+      "step": 2150
+    },
+    {
+      "epoch": 2.0175110903572264,
+      "grad_norm": 0.03431848280739808,
+      "learning_rate": 0.00010671698896559968,
+      "loss": 1.5599,
+      "step": 2160
+    },
+    {
+      "epoch": 2.0268503385477468,
+      "grad_norm": 0.04234332973849956,
+      "learning_rate": 0.00010594589167966606,
+      "loss": 1.5494,
+      "step": 2170
+    },
+    {
+      "epoch": 2.0361895867382676,
+      "grad_norm": 0.03260321438171042,
+      "learning_rate": 0.00010517443943339438,
+      "loss": 1.5473,
+      "step": 2180
+    },
+    {
+      "epoch": 2.0455288349287883,
+      "grad_norm": 0.034116901745609114,
+      "learning_rate": 0.00010440267828126478,
+      "loss": 1.5464,
+      "step": 2190
+    },
+    {
+      "epoch": 2.0548680831193087,
+      "grad_norm": 0.030992757239375807,
+      "learning_rate": 0.00010363065429619858,
+      "loss": 1.5514,
+      "step": 2200
+    },
+    {
+      "epoch": 2.0642073313098295,
+      "grad_norm": 0.03365516197786113,
+      "learning_rate": 0.0001028584135668077,
+      "loss": 1.5493,
+      "step": 2210
+    },
+    {
+      "epoch": 2.0735465795003503,
+      "grad_norm": 0.033293307482261586,
+      "learning_rate": 0.00010208600219464355,
+      "loss": 1.5426,
+      "step": 2220
+    },
+    {
+      "epoch": 2.082885827690871,
+      "grad_norm": 0.03653097737467338,
+      "learning_rate": 0.00010131346629144451,
+      "loss": 1.5471,
+      "step": 2230
+    },
+    {
+      "epoch": 2.0922250758813914,
+      "grad_norm": 0.03390291511697895,
+      "learning_rate": 0.0001005408519763833,
+      "loss": 1.5568,
+      "step": 2240
+    },
+    {
+      "epoch": 2.101564324071912,
+      "grad_norm": 0.03192694661852283,
+      "learning_rate": 9.976820537331374e-05,
+      "loss": 1.5452,
+      "step": 2250
+    },
+    {
+      "epoch": 2.110903572262433,
+      "grad_norm": 0.03561740193515691,
+      "learning_rate": 9.899557260801707e-05,
+      "loss": 1.546,
+      "step": 2260
+    },
+    {
+      "epoch": 2.1202428204529538,
+      "grad_norm": 0.029803732953068658,
+      "learning_rate": 9.822299980544862e-05,
+      "loss": 1.5533,
+      "step": 2270
+    },
+    {
+      "epoch": 2.129582068643474,
+      "grad_norm": 0.031232417271289125,
+      "learning_rate": 9.745053308698392e-05,
+      "loss": 1.5469,
+      "step": 2280
+    },
+    {
+      "epoch": 2.138921316833995,
+      "grad_norm": 0.032434793780181034,
+      "learning_rate": 9.667821856766548e-05,
+      "loss": 1.5514,
+      "step": 2290
+    },
+    {
+      "epoch": 2.1482605650245157,
+      "grad_norm": 0.03579370906405582,
+      "learning_rate": 9.590610235344972e-05,
+      "loss": 1.5577,
+      "step": 2300
+    },
+    {
+      "epoch": 2.157599813215036,
+      "grad_norm": 0.029662202478648328,
+      "learning_rate": 9.51342305384546e-05,
+      "loss": 1.5543,
+      "step": 2310
+    },
+    {
+      "epoch": 2.166939061405557,
+      "grad_norm": 0.03178715913934592,
+      "learning_rate": 9.436264920220781e-05,
+      "loss": 1.5579,
+      "step": 2320
+    },
+    {
+      "epoch": 2.1762783095960776,
+      "grad_norm": 0.03384008887051677,
+      "learning_rate": 9.359140440689601e-05,
+      "loss": 1.5595,
+      "step": 2330
+    },
+    {
+      "epoch": 2.1856175577865984,
+      "grad_norm": 0.03316450664408166,
+      "learning_rate": 9.282054219461475e-05,
+      "loss": 1.5556,
+      "step": 2340
+    },
+    {
+      "epoch": 2.1949568059771187,
+      "grad_norm": 0.032176305552558876,
+      "learning_rate": 9.205010858462007e-05,
+      "loss": 1.5638,
+      "step": 2350
+    },
+    {
+      "epoch": 2.2042960541676395,
+      "grad_norm": 0.031134335256756362,
+      "learning_rate": 9.128014957058109e-05,
+      "loss": 1.5629,
+      "step": 2360
+    },
+    {
+      "epoch": 2.2136353023581603,
+      "grad_norm": 0.032205851810441756,
+      "learning_rate": 9.051071111783436e-05,
+      "loss": 1.5613,
+      "step": 2370
+    },
+    {
+      "epoch": 2.2229745505486807,
+      "grad_norm": 0.029048245476020442,
+      "learning_rate": 8.974183916063968e-05,
+      "loss": 1.5594,
+      "step": 2380
+    },
+    {
+      "epoch": 2.2323137987392014,
+      "grad_norm": 0.03183166054573621,
+      "learning_rate": 8.897357959943795e-05,
+      "loss": 1.5606,
+      "step": 2390
+    },
+    {
+      "epoch": 2.2416530469297222,
+      "grad_norm": 0.032309922117136916,
+      "learning_rate": 8.820597829811109e-05,
+      "loss": 1.5524,
+      "step": 2400
+    },
+    {
+      "epoch": 2.250992295120243,
+      "grad_norm": 0.03598922231958808,
+      "learning_rate": 8.743908108124388e-05,
+      "loss": 1.5604,
+      "step": 2410
+    },
+    {
+      "epoch": 2.2603315433107634,
+      "grad_norm": 0.03046424735786346,
+      "learning_rate": 8.667293373138835e-05,
+      "loss": 1.5598,
+      "step": 2420
+    },
+    {
+      "epoch": 2.269670791501284,
+      "grad_norm": 0.030995453538377543,
+      "learning_rate": 8.59075819863307e-05,
+      "loss": 1.5652,
+      "step": 2430
+    },
+    {
+      "epoch": 2.279010039691805,
+      "grad_norm": 0.029309020623010097,
+      "learning_rate": 8.514307153636077e-05,
+      "loss": 1.5651,
+      "step": 2440
+    },
+    {
+      "epoch": 2.2883492878823253,
+      "grad_norm": 0.03158721106736763,
+      "learning_rate": 8.437944802154434e-05,
+      "loss": 1.5581,
+      "step": 2450
+    },
+    {
+      "epoch": 2.297688536072846,
+      "grad_norm": 0.03168229084049938,
+      "learning_rate": 8.361675702899871e-05,
+      "loss": 1.5671,
+      "step": 2460
+    },
+    {
+      "epoch": 2.307027784263367,
+      "grad_norm": 0.031335222148495136,
+      "learning_rate": 8.2855044090171e-05,
+      "loss": 1.5675,
+      "step": 2470
+    },
+    {
+      "epoch": 2.3163670324538876,
+      "grad_norm": 0.031073847325941303,
+      "learning_rate": 8.209435467811998e-05,
+      "loss": 1.5624,
+      "step": 2480
+    },
+    {
+      "epoch": 2.325706280644408,
+      "grad_norm": 0.030099100631045844,
+      "learning_rate": 8.133473420480161e-05,
+      "loss": 1.5606,
+      "step": 2490
+    },
+    {
+      "epoch": 2.3350455288349288,
+      "grad_norm": 0.034147632950361176,
+      "learning_rate": 8.057622801835788e-05,
+      "loss": 1.5703,
+      "step": 2500
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 4280,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 100,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.0242534775648158e+20,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

l2-13b-ga/checkpoint-2500/zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

l2-13b-ga/checkpoint-3700/config.json

@@ -0,0 +1,30 @@
+{
+  "_name_or_path": "meta-llama/Llama-2-13b-hf",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 5120,
+  "initializer_range": 0.02,
+  "intermediate_size": 13824,
+  "max_position_embeddings": 4096,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 40,
+  "num_hidden_layers": 40,
+  "num_key_value_heads": 40,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.46.3",
+  "use_cache": true,
+  "vocab_size": 35483
+}

l2-13b-ga/checkpoint-3700/generation_config.json

@@ -0,0 +1,10 @@
+{
+  "bos_token_id": 1,
+  "do_sample": true,
+  "eos_token_id": 2,
+  "max_length": 4096,
+  "pad_token_id": 0,
+  "temperature": 0.6,
+  "top_p": 0.9,
+  "transformers_version": "4.46.3"
+}

l2-13b-ga/checkpoint-3700/latest

@@ -0,0 +1 @@
+global_step3700

l2-13b-ga/checkpoint-3700/model.safetensors.index.json

@@ -0,0 +1,370 @@
+{
+  "metadata": {
+    "total_size": 26103060480
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00006-of-00006.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.15.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.30.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.31.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.input_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.down_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.gate_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.up_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.post_attention_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.39.input_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.down_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.gate_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.up_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.post_attention_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.k_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.o_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.q_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.v_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.norm.weight": "model-00006-of-00006.safetensors"
+  }
+}

l2-13b-ga/checkpoint-3700/special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

l2-13b-ga/checkpoint-3700/tokenizer_config.json

@@ -0,0 +1,42 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": true,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": true,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

l2-13b-ga/checkpoint-3700/trainer_state.json

@@ -0,0 +1,2630 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.4558720522997897,
+  "eval_steps": 500,
+  "global_step": 3700,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0009339248190520663,
+      "grad_norm": 6.638877692627699,
+      "learning_rate": 9.345794392523364e-07,
+      "loss": 9.2917,
+      "step": 1
+    },
+    {
+      "epoch": 0.009339248190520663,
+      "grad_norm": 1.1560921335705272,
+      "learning_rate": 9.345794392523365e-06,
+      "loss": 9.0876,
+      "step": 10
+    },
+    {
+      "epoch": 0.018678496381041326,
+      "grad_norm": 0.8415132296956432,
+      "learning_rate": 1.869158878504673e-05,
+      "loss": 8.2164,
+      "step": 20
+    },
+    {
+      "epoch": 0.02801774457156199,
+      "grad_norm": 0.45381630992958155,
+      "learning_rate": 2.8037383177570094e-05,
+      "loss": 7.5184,
+      "step": 30
+    },
+    {
+      "epoch": 0.03735699276208265,
+      "grad_norm": 0.8400636107958425,
+      "learning_rate": 3.738317757009346e-05,
+      "loss": 6.6507,
+      "step": 40
+    },
+    {
+      "epoch": 0.046696240952603316,
+      "grad_norm": 0.557696240829066,
+      "learning_rate": 4.672897196261683e-05,
+      "loss": 5.8909,
+      "step": 50
+    },
+    {
+      "epoch": 0.05603548914312398,
+      "grad_norm": 0.3971996057467842,
+      "learning_rate": 5.607476635514019e-05,
+      "loss": 5.4127,
+      "step": 60
+    },
+    {
+      "epoch": 0.06537473733364464,
+      "grad_norm": 0.2932710540265688,
+      "learning_rate": 6.542056074766355e-05,
+      "loss": 5.0106,
+      "step": 70
+    },
+    {
+      "epoch": 0.0747139855241653,
+      "grad_norm": 0.3682690443551033,
+      "learning_rate": 7.476635514018692e-05,
+      "loss": 4.6042,
+      "step": 80
+    },
+    {
+      "epoch": 0.08405323371468597,
+      "grad_norm": 0.3132971920011515,
+      "learning_rate": 8.411214953271028e-05,
+      "loss": 4.2031,
+      "step": 90
+    },
+    {
+      "epoch": 0.09339248190520663,
+      "grad_norm": 0.6731868159213446,
+      "learning_rate": 9.345794392523365e-05,
+      "loss": 3.9423,
+      "step": 100
+    },
+    {
+      "epoch": 0.1027317300957273,
+      "grad_norm": 0.27848867836763197,
+      "learning_rate": 0.000102803738317757,
+      "loss": 3.7157,
+      "step": 110
+    },
+    {
+      "epoch": 0.11207097828624796,
+      "grad_norm": 0.24642109032991807,
+      "learning_rate": 0.00011214953271028037,
+      "loss": 3.4516,
+      "step": 120
+    },
+    {
+      "epoch": 0.12141022647676862,
+      "grad_norm": 0.25717384664029797,
+      "learning_rate": 0.00012149532710280373,
+      "loss": 3.2167,
+      "step": 130
+    },
+    {
+      "epoch": 0.13074947466728928,
+      "grad_norm": 0.20912922668565637,
+      "learning_rate": 0.0001308411214953271,
+      "loss": 3.0237,
+      "step": 140
+    },
+    {
+      "epoch": 0.14008872285780993,
+      "grad_norm": 0.15805888388706113,
+      "learning_rate": 0.00014018691588785047,
+      "loss": 2.8529,
+      "step": 150
+    },
+    {
+      "epoch": 0.1494279710483306,
+      "grad_norm": 0.23370349497479534,
+      "learning_rate": 0.00014953271028037384,
+      "loss": 2.7078,
+      "step": 160
+    },
+    {
+      "epoch": 0.15876721923885126,
+      "grad_norm": 0.1802138633012483,
+      "learning_rate": 0.0001588785046728972,
+      "loss": 2.6115,
+      "step": 170
+    },
+    {
+      "epoch": 0.16810646742937194,
+      "grad_norm": 0.13354347610039718,
+      "learning_rate": 0.00016822429906542056,
+      "loss": 2.5309,
+      "step": 180
+    },
+    {
+      "epoch": 0.17744571561989259,
+      "grad_norm": 0.09414865188086892,
+      "learning_rate": 0.00017757009345794393,
+      "loss": 2.4452,
+      "step": 190
+    },
+    {
+      "epoch": 0.18678496381041326,
+      "grad_norm": 0.08333601554768896,
+      "learning_rate": 0.0001869158878504673,
+      "loss": 2.3832,
+      "step": 200
+    },
+    {
+      "epoch": 0.1961242120009339,
+      "grad_norm": 0.15926414699806835,
+      "learning_rate": 0.00019626168224299065,
+      "loss": 2.3492,
+      "step": 210
+    },
+    {
+      "epoch": 0.2054634601914546,
+      "grad_norm": 0.09492820761057012,
+      "learning_rate": 0.0001999989254250208,
+      "loss": 2.323,
+      "step": 220
+    },
+    {
+      "epoch": 0.21480270838197524,
+      "grad_norm": 0.0801349259356147,
+      "learning_rate": 0.00019999235866155886,
+      "loss": 2.2731,
+      "step": 230
+    },
+    {
+      "epoch": 0.22414195657249592,
+      "grad_norm": 0.12210960524693895,
+      "learning_rate": 0.00019997982251228469,
+      "loss": 2.2433,
+      "step": 240
+    },
+    {
+      "epoch": 0.23348120476301656,
+      "grad_norm": 3.14289498732125,
+      "learning_rate": 0.00019996131772558666,
+      "loss": 3.2769,
+      "step": 250
+    },
+    {
+      "epoch": 0.24282045295353724,
+      "grad_norm": 1.632940983166179,
+      "learning_rate": 0.00019993684540617132,
+      "loss": 4.9343,
+      "step": 260
+    },
+    {
+      "epoch": 0.2521597011440579,
+      "grad_norm": 3.4831252230225416,
+      "learning_rate": 0.00019990640701499736,
+      "loss": 4.2768,
+      "step": 270
+    },
+    {
+      "epoch": 0.26149894933457857,
+      "grad_norm": 1.6069045920523788,
+      "learning_rate": 0.00019987000436918874,
+      "loss": 5.9581,
+      "step": 280
+    },
+    {
+      "epoch": 0.27083819752509924,
+      "grad_norm": 0.2220907936615993,
+      "learning_rate": 0.00019982763964192585,
+      "loss": 3.8228,
+      "step": 290
+    },
+    {
+      "epoch": 0.28017744571561987,
+      "grad_norm": 0.24737284913291765,
+      "learning_rate": 0.00019977931536231596,
+      "loss": 3.1413,
+      "step": 300
+    },
+    {
+      "epoch": 0.28951669390614054,
+      "grad_norm": 4.010404518241152,
+      "learning_rate": 0.00019972503441524224,
+      "loss": 2.8432,
+      "step": 310
+    },
+    {
+      "epoch": 0.2988559420966612,
+      "grad_norm": 0.1515583580811596,
+      "learning_rate": 0.00019966480004119142,
+      "loss": 2.7859,
+      "step": 320
+    },
+    {
+      "epoch": 0.3081951902871819,
+      "grad_norm": 0.11259395750650594,
+      "learning_rate": 0.00019959861583606045,
+      "loss": 2.5821,
+      "step": 330
+    },
+    {
+      "epoch": 0.3175344384777025,
+      "grad_norm": 0.22514797814956813,
+      "learning_rate": 0.00019952648575094183,
+      "loss": 2.4517,
+      "step": 340
+    },
+    {
+      "epoch": 0.3268736866682232,
+      "grad_norm": 0.08040136172033542,
+      "learning_rate": 0.00019944841409188767,
+      "loss": 2.3794,
+      "step": 350
+    },
+    {
+      "epoch": 0.3362129348587439,
+      "grad_norm": 0.054758073593565354,
+      "learning_rate": 0.00019936440551965263,
+      "loss": 2.3232,
+      "step": 360
+    },
+    {
+      "epoch": 0.34555218304926455,
+      "grad_norm": 0.06742998909645591,
+      "learning_rate": 0.00019927446504941577,
+      "loss": 2.2776,
+      "step": 370
+    },
+    {
+      "epoch": 0.35489143123978517,
+      "grad_norm": 0.048780907584876736,
+      "learning_rate": 0.00019917859805048096,
+      "loss": 2.2376,
+      "step": 380
+    },
+    {
+      "epoch": 0.36423067943030585,
+      "grad_norm": 0.0475325963052214,
+      "learning_rate": 0.00019907681024595663,
+      "loss": 2.2191,
+      "step": 390
+    },
+    {
+      "epoch": 0.3735699276208265,
+      "grad_norm": 0.054089563211590065,
+      "learning_rate": 0.00019896910771241387,
+      "loss": 2.1961,
+      "step": 400
+    },
+    {
+      "epoch": 0.3829091758113472,
+      "grad_norm": 0.21798406131864823,
+      "learning_rate": 0.00019885549687952372,
+      "loss": 2.2078,
+      "step": 410
+    },
+    {
+      "epoch": 0.3922484240018678,
+      "grad_norm": 0.8673185709111124,
+      "learning_rate": 0.00019873598452967338,
+      "loss": 2.3731,
+      "step": 420
+    },
+    {
+      "epoch": 0.4015876721923885,
+      "grad_norm": 0.22424350669971718,
+      "learning_rate": 0.0001986105777975613,
+      "loss": 2.6195,
+      "step": 430
+    },
+    {
+      "epoch": 0.4109269203829092,
+      "grad_norm": 0.307418135168262,
+      "learning_rate": 0.00019847928416977126,
+      "loss": 2.3624,
+      "step": 440
+    },
+    {
+      "epoch": 0.42026616857342985,
+      "grad_norm": 0.07944722668080402,
+      "learning_rate": 0.00019834211148432536,
+      "loss": 2.2799,
+      "step": 450
+    },
+    {
+      "epoch": 0.4296054167639505,
+      "grad_norm": 0.18146933758664588,
+      "learning_rate": 0.00019819906793021614,
+      "loss": 2.2177,
+      "step": 460
+    },
+    {
+      "epoch": 0.43894466495447115,
+      "grad_norm": 0.07035825837333018,
+      "learning_rate": 0.0001980501620469178,
+      "loss": 2.1767,
+      "step": 470
+    },
+    {
+      "epoch": 0.44828391314499183,
+      "grad_norm": 0.04596186944454228,
+      "learning_rate": 0.0001978954027238763,
+      "loss": 2.1598,
+      "step": 480
+    },
+    {
+      "epoch": 0.4576231613355125,
+      "grad_norm": 0.041342347745088055,
+      "learning_rate": 0.0001977347991999786,
+      "loss": 2.131,
+      "step": 490
+    },
+    {
+      "epoch": 0.46696240952603313,
+      "grad_norm": 0.04172063219841485,
+      "learning_rate": 0.00019756836106300137,
+      "loss": 2.1231,
+      "step": 500
+    },
+    {
+      "epoch": 0.4763016577165538,
+      "grad_norm": 0.03373646457711144,
+      "learning_rate": 0.00019739609824903843,
+      "loss": 2.1146,
+      "step": 510
+    },
+    {
+      "epoch": 0.4856409059070745,
+      "grad_norm": 0.03736871030676605,
+      "learning_rate": 0.00019721802104190748,
+      "loss": 2.1003,
+      "step": 520
+    },
+    {
+      "epoch": 0.49498015409759516,
+      "grad_norm": 0.033931028038211034,
+      "learning_rate": 0.00019703414007253645,
+      "loss": 2.0983,
+      "step": 530
+    },
+    {
+      "epoch": 0.5043194022881158,
+      "grad_norm": 0.03790055446070549,
+      "learning_rate": 0.00019684446631832868,
+      "loss": 2.092,
+      "step": 540
+    },
+    {
+      "epoch": 0.5136586504786365,
+      "grad_norm": 0.030956192803893078,
+      "learning_rate": 0.00019664901110250758,
+      "loss": 2.0807,
+      "step": 550
+    },
+    {
+      "epoch": 0.5229978986691571,
+      "grad_norm": 0.03542530209935129,
+      "learning_rate": 0.00019644778609344068,
+      "loss": 2.0773,
+      "step": 560
+    },
+    {
+      "epoch": 0.5323371468596778,
+      "grad_norm": 0.040947757568902336,
+      "learning_rate": 0.00019624080330394306,
+      "loss": 2.0649,
+      "step": 570
+    },
+    {
+      "epoch": 0.5416763950501985,
+      "grad_norm": 0.034273415973688146,
+      "learning_rate": 0.00019602807509056018,
+      "loss": 2.0479,
+      "step": 580
+    },
+    {
+      "epoch": 0.5510156432407192,
+      "grad_norm": 0.031427481498873144,
+      "learning_rate": 0.00019580961415283028,
+      "loss": 2.0563,
+      "step": 590
+    },
+    {
+      "epoch": 0.5603548914312397,
+      "grad_norm": 0.03141549752041532,
+      "learning_rate": 0.00019558543353252611,
+      "loss": 2.0503,
+      "step": 600
+    },
+    {
+      "epoch": 0.5696941396217604,
+      "grad_norm": 0.033012392726428204,
+      "learning_rate": 0.00019535554661287652,
+      "loss": 2.0389,
+      "step": 610
+    },
+    {
+      "epoch": 0.5790333878122811,
+      "grad_norm": 0.02913261992661444,
+      "learning_rate": 0.0001951199671177673,
+      "loss": 2.036,
+      "step": 620
+    },
+    {
+      "epoch": 0.5883726360028018,
+      "grad_norm": 0.030543903708435332,
+      "learning_rate": 0.00019487870911092214,
+      "loss": 2.0326,
+      "step": 630
+    },
+    {
+      "epoch": 0.5977118841933224,
+      "grad_norm": 0.03215005545393897,
+      "learning_rate": 0.00019463178699506277,
+      "loss": 2.0231,
+      "step": 640
+    },
+    {
+      "epoch": 0.6070511323838431,
+      "grad_norm": 0.03823630791937631,
+      "learning_rate": 0.00019437921551104933,
+      "loss": 2.0293,
+      "step": 650
+    },
+    {
+      "epoch": 0.6163903805743638,
+      "grad_norm": 0.03200103149471209,
+      "learning_rate": 0.00019412100973700038,
+      "loss": 2.017,
+      "step": 660
+    },
+    {
+      "epoch": 0.6257296287648845,
+      "grad_norm": 0.03841804153577787,
+      "learning_rate": 0.00019385718508739262,
+      "loss": 2.0135,
+      "step": 670
+    },
+    {
+      "epoch": 0.635068876955405,
+      "grad_norm": 0.03052396655271533,
+      "learning_rate": 0.0001935877573121407,
+      "loss": 2.0237,
+      "step": 680
+    },
+    {
+      "epoch": 0.6444081251459257,
+      "grad_norm": 0.033817837533771815,
+      "learning_rate": 0.00019331274249565717,
+      "loss": 2.0069,
+      "step": 690
+    },
+    {
+      "epoch": 0.6537473733364464,
+      "grad_norm": 0.028286999650643876,
+      "learning_rate": 0.00019303215705589194,
+      "loss": 2.0112,
+      "step": 700
+    },
+    {
+      "epoch": 0.6630866215269671,
+      "grad_norm": 0.030693948626257357,
+      "learning_rate": 0.00019274601774335243,
+      "loss": 2.0,
+      "step": 710
+    },
+    {
+      "epoch": 0.6724258697174877,
+      "grad_norm": 0.029984615635953022,
+      "learning_rate": 0.0001924543416401035,
+      "loss": 2.0028,
+      "step": 720
+    },
+    {
+      "epoch": 0.6817651179080084,
+      "grad_norm": 0.02915985673921391,
+      "learning_rate": 0.00019215714615874755,
+      "loss": 2.0031,
+      "step": 730
+    },
+    {
+      "epoch": 0.6911043660985291,
+      "grad_norm": 0.028305399777245336,
+      "learning_rate": 0.00019185444904138528,
+      "loss": 1.9924,
+      "step": 740
+    },
+    {
+      "epoch": 0.7004436142890498,
+      "grad_norm": 0.036720505429756495,
+      "learning_rate": 0.00019154626835855628,
+      "loss": 1.9981,
+      "step": 750
+    },
+    {
+      "epoch": 0.7097828624795703,
+      "grad_norm": 0.0287124048917296,
+      "learning_rate": 0.00019123262250816034,
+      "loss": 1.9868,
+      "step": 760
+    },
+    {
+      "epoch": 0.719122110670091,
+      "grad_norm": 0.03318092492837997,
+      "learning_rate": 0.00019091353021435915,
+      "loss": 1.9943,
+      "step": 770
+    },
+    {
+      "epoch": 0.7284613588606117,
+      "grad_norm": 0.054035272137015325,
+      "learning_rate": 0.00019058901052645844,
+      "loss": 1.9838,
+      "step": 780
+    },
+    {
+      "epoch": 0.7378006070511324,
+      "grad_norm": 0.03184392761983255,
+      "learning_rate": 0.00019025908281777078,
+      "loss": 1.982,
+      "step": 790
+    },
+    {
+      "epoch": 0.747139855241653,
+      "grad_norm": 0.029409948164434735,
+      "learning_rate": 0.00018992376678445908,
+      "loss": 1.9693,
+      "step": 800
+    },
+    {
+      "epoch": 0.7564791034321737,
+      "grad_norm": 0.029656963043919016,
+      "learning_rate": 0.00018958308244436064,
+      "loss": 1.9914,
+      "step": 810
+    },
+    {
+      "epoch": 0.7658183516226944,
+      "grad_norm": 0.030843610865326686,
+      "learning_rate": 0.00018923705013579233,
+      "loss": 1.9749,
+      "step": 820
+    },
+    {
+      "epoch": 0.7751575998132151,
+      "grad_norm": 0.0377584286045999,
+      "learning_rate": 0.00018888569051633613,
+      "loss": 1.9606,
+      "step": 830
+    },
+    {
+      "epoch": 0.7844968480037356,
+      "grad_norm": 0.03147699916274391,
+      "learning_rate": 0.00018852902456160616,
+      "loss": 1.9696,
+      "step": 840
+    },
+    {
+      "epoch": 0.7938360961942563,
+      "grad_norm": 0.029193153251471263,
+      "learning_rate": 0.0001881670735639963,
+      "loss": 1.9687,
+      "step": 850
+    },
+    {
+      "epoch": 0.803175344384777,
+      "grad_norm": 0.030201895228156087,
+      "learning_rate": 0.00018779985913140924,
+      "loss": 1.9678,
+      "step": 860
+    },
+    {
+      "epoch": 0.8125145925752977,
+      "grad_norm": 0.029272319485493213,
+      "learning_rate": 0.00018742740318596632,
+      "loss": 1.9697,
+      "step": 870
+    },
+    {
+      "epoch": 0.8218538407658184,
+      "grad_norm": 0.033740176465285654,
+      "learning_rate": 0.000187049727962699,
+      "loss": 1.9647,
+      "step": 880
+    },
+    {
+      "epoch": 0.831193088956339,
+      "grad_norm": 0.029539399251208593,
+      "learning_rate": 0.0001866668560082213,
+      "loss": 1.9627,
+      "step": 890
+    },
+    {
+      "epoch": 0.8405323371468597,
+      "grad_norm": 0.028666077337951026,
+      "learning_rate": 0.0001862788101793839,
+      "loss": 1.9529,
+      "step": 900
+    },
+    {
+      "epoch": 0.8498715853373804,
+      "grad_norm": 0.030280792492665805,
+      "learning_rate": 0.0001858856136419097,
+      "loss": 1.9536,
+      "step": 910
+    },
+    {
+      "epoch": 0.859210833527901,
+      "grad_norm": 0.03921536105057096,
+      "learning_rate": 0.0001854872898690106,
+      "loss": 1.9474,
+      "step": 920
+    },
+    {
+      "epoch": 0.8685500817184216,
+      "grad_norm": 0.030632523637038354,
+      "learning_rate": 0.0001850838626399865,
+      "loss": 1.9423,
+      "step": 930
+    },
+    {
+      "epoch": 0.8778893299089423,
+      "grad_norm": 0.04615147601979514,
+      "learning_rate": 0.00018467535603880548,
+      "loss": 1.946,
+      "step": 940
+    },
+    {
+      "epoch": 0.887228578099463,
+      "grad_norm": 0.028216236017006333,
+      "learning_rate": 0.00018426179445266616,
+      "loss": 1.9408,
+      "step": 950
+    },
+    {
+      "epoch": 0.8965678262899837,
+      "grad_norm": 0.0282407563402959,
+      "learning_rate": 0.00018384320257054177,
+      "loss": 1.9447,
+      "step": 960
+    },
+    {
+      "epoch": 0.9059070744805043,
+      "grad_norm": 0.029365880854687894,
+      "learning_rate": 0.0001834196053817062,
+      "loss": 1.9389,
+      "step": 970
+    },
+    {
+      "epoch": 0.915246322671025,
+      "grad_norm": 0.02855404439130719,
+      "learning_rate": 0.00018299102817424234,
+      "loss": 1.9425,
+      "step": 980
+    },
+    {
+      "epoch": 0.9245855708615457,
+      "grad_norm": 0.03238310984070135,
+      "learning_rate": 0.00018255749653353225,
+      "loss": 1.9392,
+      "step": 990
+    },
+    {
+      "epoch": 0.9339248190520663,
+      "grad_norm": 0.030115350805430388,
+      "learning_rate": 0.00018211903634072983,
+      "loss": 1.956,
+      "step": 1000
+    },
+    {
+      "epoch": 0.9432640672425869,
+      "grad_norm": 0.039090564032501135,
+      "learning_rate": 0.0001816756737712158,
+      "loss": 1.9358,
+      "step": 1010
+    },
+    {
+      "epoch": 0.9526033154331076,
+      "grad_norm": 0.028567954476327023,
+      "learning_rate": 0.000181227435293035,
+      "loss": 1.9342,
+      "step": 1020
+    },
+    {
+      "epoch": 0.9619425636236283,
+      "grad_norm": 0.028158653241284505,
+      "learning_rate": 0.00018077434766531624,
+      "loss": 1.9287,
+      "step": 1030
+    },
+    {
+      "epoch": 0.971281811814149,
+      "grad_norm": 0.02927795717651538,
+      "learning_rate": 0.00018031643793667504,
+      "loss": 1.9298,
+      "step": 1040
+    },
+    {
+      "epoch": 0.9806210600046696,
+      "grad_norm": 0.06298340584032344,
+      "learning_rate": 0.0001798537334435986,
+      "loss": 1.9303,
+      "step": 1050
+    },
+    {
+      "epoch": 0.9899603081951903,
+      "grad_norm": 0.03627278406983255,
+      "learning_rate": 0.00017938626180881407,
+      "loss": 1.9285,
+      "step": 1060
+    },
+    {
+      "epoch": 0.999299556385711,
+      "grad_norm": 0.03916823421329747,
+      "learning_rate": 0.00017891405093963938,
+      "loss": 1.9239,
+      "step": 1070
+    },
+    {
+      "epoch": 1.0087555451786132,
+      "grad_norm": 0.03383539251115568,
+      "learning_rate": 0.00017843712902631723,
+      "loss": 1.8855,
+      "step": 1080
+    },
+    {
+      "epoch": 1.0180947933691338,
+      "grad_norm": 0.02931021936560147,
+      "learning_rate": 0.00017795552454033224,
+      "loss": 1.8004,
+      "step": 1090
+    },
+    {
+      "epoch": 1.0274340415596543,
+      "grad_norm": 0.029173295095350292,
+      "learning_rate": 0.0001774692662327113,
+      "loss": 1.7912,
+      "step": 1100
+    },
+    {
+      "epoch": 1.0367732897501751,
+      "grad_norm": 0.035210065642974735,
+      "learning_rate": 0.000176978383132307,
+      "loss": 1.7902,
+      "step": 1110
+    },
+    {
+      "epoch": 1.0461125379406957,
+      "grad_norm": 0.029794447659573477,
+      "learning_rate": 0.00017648290454406475,
+      "loss": 1.8072,
+      "step": 1120
+    },
+    {
+      "epoch": 1.0554517861312165,
+      "grad_norm": 0.03194584938279939,
+      "learning_rate": 0.0001759828600472734,
+      "loss": 1.803,
+      "step": 1130
+    },
+    {
+      "epoch": 1.064791034321737,
+      "grad_norm": 0.031028415312581603,
+      "learning_rate": 0.00017547827949379924,
+      "loss": 1.7945,
+      "step": 1140
+    },
+    {
+      "epoch": 1.0741302825122578,
+      "grad_norm": 0.03313245828751237,
+      "learning_rate": 0.00017496919300630403,
+      "loss": 1.8139,
+      "step": 1150
+    },
+    {
+      "epoch": 1.0834695307027784,
+      "grad_norm": 0.03049431724979126,
+      "learning_rate": 0.00017445563097644664,
+      "loss": 1.8031,
+      "step": 1160
+    },
+    {
+      "epoch": 1.0928087788932992,
+      "grad_norm": 0.02880548705343715,
+      "learning_rate": 0.00017393762406306878,
+      "loss": 1.7974,
+      "step": 1170
+    },
+    {
+      "epoch": 1.1021480270838198,
+      "grad_norm": 0.03200427895977668,
+      "learning_rate": 0.00017341520319036469,
+      "loss": 1.7994,
+      "step": 1180
+    },
+    {
+      "epoch": 1.1114872752743403,
+      "grad_norm": 0.031293532815600045,
+      "learning_rate": 0.00017288839954603496,
+      "loss": 1.8073,
+      "step": 1190
+    },
+    {
+      "epoch": 1.1208265234648611,
+      "grad_norm": 0.032367211109345505,
+      "learning_rate": 0.00017235724457942468,
+      "loss": 1.7944,
+      "step": 1200
+    },
+    {
+      "epoch": 1.1301657716553817,
+      "grad_norm": 0.037783793923191374,
+      "learning_rate": 0.0001718217699996462,
+      "loss": 1.7948,
+      "step": 1210
+    },
+    {
+      "epoch": 1.1395050198459025,
+      "grad_norm": 0.02881083117349317,
+      "learning_rate": 0.00017128200777368567,
+      "loss": 1.8029,
+      "step": 1220
+    },
+    {
+      "epoch": 1.148844268036423,
+      "grad_norm": 0.032997408408175985,
+      "learning_rate": 0.00017073799012449524,
+      "loss": 1.7914,
+      "step": 1230
+    },
+    {
+      "epoch": 1.1581835162269438,
+      "grad_norm": 0.030263615801190885,
+      "learning_rate": 0.00017018974952906884,
+      "loss": 1.792,
+      "step": 1240
+    },
+    {
+      "epoch": 1.1675227644174644,
+      "grad_norm": 0.03286310429098484,
+      "learning_rate": 0.00016963731871650378,
+      "loss": 1.8149,
+      "step": 1250
+    },
+    {
+      "epoch": 1.1768620126079852,
+      "grad_norm": 0.031111621587597126,
+      "learning_rate": 0.00016908073066604663,
+      "loss": 1.8092,
+      "step": 1260
+    },
+    {
+      "epoch": 1.1862012607985057,
+      "grad_norm": 0.03372950172075241,
+      "learning_rate": 0.0001685200186051246,
+      "loss": 1.818,
+      "step": 1270
+    },
+    {
+      "epoch": 1.1955405089890263,
+      "grad_norm": 0.040782040436567434,
+      "learning_rate": 0.00016795521600736164,
+      "loss": 1.7999,
+      "step": 1280
+    },
+    {
+      "epoch": 1.204879757179547,
+      "grad_norm": 0.03451605168178924,
+      "learning_rate": 0.00016738635659058044,
+      "loss": 1.7945,
+      "step": 1290
+    },
+    {
+      "epoch": 1.2142190053700677,
+      "grad_norm": 0.03235681588882673,
+      "learning_rate": 0.00016681347431478933,
+      "loss": 1.8087,
+      "step": 1300
+    },
+    {
+      "epoch": 1.2235582535605884,
+      "grad_norm": 0.030750745605971932,
+      "learning_rate": 0.00016623660338015487,
+      "loss": 1.7995,
+      "step": 1310
+    },
+    {
+      "epoch": 1.232897501751109,
+      "grad_norm": 0.029444668665577274,
+      "learning_rate": 0.00016565577822496042,
+      "loss": 1.8025,
+      "step": 1320
+    },
+    {
+      "epoch": 1.2422367499416298,
+      "grad_norm": 0.038528856709584745,
+      "learning_rate": 0.00016507103352354996,
+      "loss": 1.7954,
+      "step": 1330
+    },
+    {
+      "epoch": 1.2515759981321504,
+      "grad_norm": 0.034217088004383035,
+      "learning_rate": 0.00016448240418425814,
+      "loss": 1.7962,
+      "step": 1340
+    },
+    {
+      "epoch": 1.260915246322671,
+      "grad_norm": 0.030205405393195585,
+      "learning_rate": 0.00016388992534732645,
+      "loss": 1.7973,
+      "step": 1350
+    },
+    {
+      "epoch": 1.2702544945131917,
+      "grad_norm": 0.029082218516562994,
+      "learning_rate": 0.00016329363238280528,
+      "loss": 1.796,
+      "step": 1360
+    },
+    {
+      "epoch": 1.2795937427037123,
+      "grad_norm": 0.029003887688766505,
+      "learning_rate": 0.00016269356088844238,
+      "loss": 1.7946,
+      "step": 1370
+    },
+    {
+      "epoch": 1.288932990894233,
+      "grad_norm": 0.03341157363649238,
+      "learning_rate": 0.00016208974668755779,
+      "loss": 1.7972,
+      "step": 1380
+    },
+    {
+      "epoch": 1.2982722390847536,
+      "grad_norm": 0.030614480844663026,
+      "learning_rate": 0.00016148222582690517,
+      "loss": 1.7973,
+      "step": 1390
+    },
+    {
+      "epoch": 1.3076114872752744,
+      "grad_norm": 0.029741346740467405,
+      "learning_rate": 0.00016087103457452,
+      "loss": 1.8076,
+      "step": 1400
+    },
+    {
+      "epoch": 1.316950735465795,
+      "grad_norm": 0.029569313554185597,
+      "learning_rate": 0.00016025620941755424,
+      "loss": 1.8043,
+      "step": 1410
+    },
+    {
+      "epoch": 1.3262899836563156,
+      "grad_norm": 0.02947637404374054,
+      "learning_rate": 0.0001596377870600983,
+      "loss": 1.797,
+      "step": 1420
+    },
+    {
+      "epoch": 1.3356292318468364,
+      "grad_norm": 0.031005062093959545,
+      "learning_rate": 0.00015901580442098968,
+      "loss": 1.8086,
+      "step": 1430
+    },
+    {
+      "epoch": 1.344968480037357,
+      "grad_norm": 0.029493792984873927,
+      "learning_rate": 0.00015839029863160922,
+      "loss": 1.8026,
+      "step": 1440
+    },
+    {
+      "epoch": 1.3543077282278777,
+      "grad_norm": 0.0288068155951218,
+      "learning_rate": 0.0001577613070336641,
+      "loss": 1.7951,
+      "step": 1450
+    },
+    {
+      "epoch": 1.3636469764183983,
+      "grad_norm": 0.03380404824627639,
+      "learning_rate": 0.00015712886717695885,
+      "loss": 1.7938,
+      "step": 1460
+    },
+    {
+      "epoch": 1.372986224608919,
+      "grad_norm": 0.039744706189693335,
+      "learning_rate": 0.0001564930168171536,
+      "loss": 1.8016,
+      "step": 1470
+    },
+    {
+      "epoch": 1.3823254727994396,
+      "grad_norm": 0.030565530594285437,
+      "learning_rate": 0.00015585379391351012,
+      "loss": 1.7984,
+      "step": 1480
+    },
+    {
+      "epoch": 1.3916647209899602,
+      "grad_norm": 0.04009392805554255,
+      "learning_rate": 0.00015521123662662567,
+      "loss": 1.7999,
+      "step": 1490
+    },
+    {
+      "epoch": 1.401003969180481,
+      "grad_norm": 0.03516196009586836,
+      "learning_rate": 0.000154565383316155,
+      "loss": 1.7979,
+      "step": 1500
+    },
+    {
+      "epoch": 1.4103432173710018,
+      "grad_norm": 0.03534161399054556,
+      "learning_rate": 0.0001539162725385202,
+      "loss": 1.8057,
+      "step": 1510
+    },
+    {
+      "epoch": 1.4196824655615223,
+      "grad_norm": 0.028488879438601067,
+      "learning_rate": 0.000153263943044609,
+      "loss": 1.792,
+      "step": 1520
+    },
+    {
+      "epoch": 1.429021713752043,
+      "grad_norm": 0.03125154490954804,
+      "learning_rate": 0.00015260843377746147,
+      "loss": 1.8008,
+      "step": 1530
+    },
+    {
+      "epoch": 1.4383609619425637,
+      "grad_norm": 0.030194357488801882,
+      "learning_rate": 0.00015194978386994507,
+      "loss": 1.7948,
+      "step": 1540
+    },
+    {
+      "epoch": 1.4477002101330843,
+      "grad_norm": 0.03049246845786265,
+      "learning_rate": 0.00015128803264241852,
+      "loss": 1.7967,
+      "step": 1550
+    },
+    {
+      "epoch": 1.4570394583236048,
+      "grad_norm": 0.030497211097258083,
+      "learning_rate": 0.0001506232196003844,
+      "loss": 1.7894,
+      "step": 1560
+    },
+    {
+      "epoch": 1.4663787065141256,
+      "grad_norm": 0.028748806119737205,
+      "learning_rate": 0.00014995538443213094,
+      "loss": 1.806,
+      "step": 1570
+    },
+    {
+      "epoch": 1.4757179547046464,
+      "grad_norm": 0.036423750322912396,
+      "learning_rate": 0.00014928456700636237,
+      "loss": 1.7995,
+      "step": 1580
+    },
+    {
+      "epoch": 1.485057202895167,
+      "grad_norm": 0.039101516109204065,
+      "learning_rate": 0.00014861080736981906,
+      "loss": 1.8028,
+      "step": 1590
+    },
+    {
+      "epoch": 1.4943964510856875,
+      "grad_norm": 0.031368399541673815,
+      "learning_rate": 0.00014793414574488663,
+      "loss": 1.8005,
+      "step": 1600
+    },
+    {
+      "epoch": 1.5037356992762083,
+      "grad_norm": 0.029788484702512056,
+      "learning_rate": 0.00014725462252719495,
+      "loss": 1.7963,
+      "step": 1610
+    },
+    {
+      "epoch": 1.5130749474667289,
+      "grad_norm": 0.029719041811636312,
+      "learning_rate": 0.00014657227828320635,
+      "loss": 1.7957,
+      "step": 1620
+    },
+    {
+      "epoch": 1.5224141956572494,
+      "grad_norm": 0.02820041575417432,
+      "learning_rate": 0.00014588715374779407,
+      "loss": 1.7986,
+      "step": 1630
+    },
+    {
+      "epoch": 1.5317534438477702,
+      "grad_norm": 1.8322544897261024,
+      "learning_rate": 0.0001451992898218102,
+      "loss": 2.016,
+      "step": 1640
+    },
+    {
+      "epoch": 1.541092692038291,
+      "grad_norm": 0.8676995793107466,
+      "learning_rate": 0.0001445087275696443,
+      "loss": 2.1095,
+      "step": 1650
+    },
+    {
+      "epoch": 1.5504319402288116,
+      "grad_norm": 0.08688193434631736,
+      "learning_rate": 0.00014381550821677155,
+      "loss": 2.0497,
+      "step": 1660
+    },
+    {
+      "epoch": 1.5597711884193322,
+      "grad_norm": 0.18321381224589608,
+      "learning_rate": 0.0001431196731472921,
+      "loss": 1.9023,
+      "step": 1670
+    },
+    {
+      "epoch": 1.569110436609853,
+      "grad_norm": 0.058214343698110564,
+      "learning_rate": 0.00014242126390145998,
+      "loss": 1.8566,
+      "step": 1680
+    },
+    {
+      "epoch": 1.5784496848003737,
+      "grad_norm": 0.03965578147557666,
+      "learning_rate": 0.0001417203221732036,
+      "loss": 1.8206,
+      "step": 1690
+    },
+    {
+      "epoch": 1.587788932990894,
+      "grad_norm": 0.03131802880017099,
+      "learning_rate": 0.00014101688980763658,
+      "loss": 1.8272,
+      "step": 1700
+    },
+    {
+      "epoch": 1.5971281811814149,
+      "grad_norm": 0.02788722706683908,
+      "learning_rate": 0.00014031100879855968,
+      "loss": 1.8145,
+      "step": 1710
+    },
+    {
+      "epoch": 1.6064674293719357,
+      "grad_norm": 0.02781057130092059,
+      "learning_rate": 0.00013960272128595372,
+      "loss": 1.8122,
+      "step": 1720
+    },
+    {
+      "epoch": 1.6158066775624562,
+      "grad_norm": 0.032067383734154756,
+      "learning_rate": 0.00013889206955346403,
+      "loss": 1.8064,
+      "step": 1730
+    },
+    {
+      "epoch": 1.6251459257529768,
+      "grad_norm": 0.03448807133884858,
+      "learning_rate": 0.00013817909602587613,
+      "loss": 1.8128,
+      "step": 1740
+    },
+    {
+      "epoch": 1.6344851739434976,
+      "grad_norm": 0.04127395288271696,
+      "learning_rate": 0.00013746384326658305,
+      "loss": 1.8041,
+      "step": 1750
+    },
+    {
+      "epoch": 1.6438244221340184,
+      "grad_norm": 0.030243192787820814,
+      "learning_rate": 0.00013674635397504427,
+      "loss": 1.803,
+      "step": 1760
+    },
+    {
+      "epoch": 1.653163670324539,
+      "grad_norm": 0.032183525625428915,
+      "learning_rate": 0.00013602667098423687,
+      "loss": 1.8092,
+      "step": 1770
+    },
+    {
+      "epoch": 1.6625029185150595,
+      "grad_norm": 0.030391044676815377,
+      "learning_rate": 0.00013530483725809818,
+      "loss": 1.8039,
+      "step": 1780
+    },
+    {
+      "epoch": 1.6718421667055803,
+      "grad_norm": 0.03204756818238517,
+      "learning_rate": 0.000134580895888961,
+      "loss": 1.8017,
+      "step": 1790
+    },
+    {
+      "epoch": 1.6811814148961008,
+      "grad_norm": 0.031117112662107062,
+      "learning_rate": 0.00013385489009498124,
+      "loss": 1.8017,
+      "step": 1800
+    },
+    {
+      "epoch": 1.6905206630866214,
+      "grad_norm": 0.028389293445401805,
+      "learning_rate": 0.00013312686321755761,
+      "loss": 1.811,
+      "step": 1810
+    },
+    {
+      "epoch": 1.6998599112771422,
+      "grad_norm": 4.908185097372493,
+      "learning_rate": 0.0001323968587187443,
+      "loss": 2.194,
+      "step": 1820
+    },
+    {
+      "epoch": 1.709199159467663,
+      "grad_norm": 0.2849452041194025,
+      "learning_rate": 0.00013166492017865637,
+      "loss": 2.0785,
+      "step": 1830
+    },
+    {
+      "epoch": 1.7185384076581836,
+      "grad_norm": 0.10989252058989733,
+      "learning_rate": 0.0001309310912928682,
+      "loss": 1.986,
+      "step": 1840
+    },
+    {
+      "epoch": 1.7278776558487041,
+      "grad_norm": 0.057475612656740484,
+      "learning_rate": 0.00013019541586980463,
+      "loss": 1.8614,
+      "step": 1850
+    },
+    {
+      "epoch": 1.737216904039225,
+      "grad_norm": 0.034908181734207726,
+      "learning_rate": 0.000129457937828126,
+      "loss": 1.8326,
+      "step": 1860
+    },
+    {
+      "epoch": 1.7465561522297455,
+      "grad_norm": 0.02892836681897248,
+      "learning_rate": 0.00012871870119410614,
+      "loss": 1.8243,
+      "step": 1870
+    },
+    {
+      "epoch": 1.755895400420266,
+      "grad_norm": 0.03311636981729384,
+      "learning_rate": 0.00012797775009900397,
+      "loss": 1.8183,
+      "step": 1880
+    },
+    {
+      "epoch": 1.7652346486107868,
+      "grad_norm": 0.03151917952358458,
+      "learning_rate": 0.00012723512877642904,
+      "loss": 1.8034,
+      "step": 1890
+    },
+    {
+      "epoch": 1.7745738968013076,
+      "grad_norm": 0.028109921832296925,
+      "learning_rate": 0.000126490881559701,
+      "loss": 1.8129,
+      "step": 1900
+    },
+    {
+      "epoch": 1.7839131449918282,
+      "grad_norm": 0.030350462454962698,
+      "learning_rate": 0.00012574505287920259,
+      "loss": 1.8003,
+      "step": 1910
+    },
+    {
+      "epoch": 1.7932523931823487,
+      "grad_norm": 0.03131380630103849,
+      "learning_rate": 0.00012499768725972754,
+      "loss": 1.814,
+      "step": 1920
+    },
+    {
+      "epoch": 1.8025916413728695,
+      "grad_norm": 0.029450198273050322,
+      "learning_rate": 0.00012424882931782243,
+      "loss": 1.7998,
+      "step": 1930
+    },
+    {
+      "epoch": 1.81193088956339,
+      "grad_norm": 0.0310524261453681,
+      "learning_rate": 0.0001234985237591231,
+      "loss": 1.8078,
+      "step": 1940
+    },
+    {
+      "epoch": 1.8212701377539107,
+      "grad_norm": 0.029362038478982822,
+      "learning_rate": 0.00012274681537568585,
+      "loss": 1.8014,
+      "step": 1950
+    },
+    {
+      "epoch": 1.8306093859444315,
+      "grad_norm": 0.027166816460226118,
+      "learning_rate": 0.00012199374904331337,
+      "loss": 1.8021,
+      "step": 1960
+    },
+    {
+      "epoch": 1.8399486341349522,
+      "grad_norm": 0.03680509283276228,
+      "learning_rate": 0.00012123936971887578,
+      "loss": 1.7973,
+      "step": 1970
+    },
+    {
+      "epoch": 1.8492878823254728,
+      "grad_norm": 0.03135174840346185,
+      "learning_rate": 0.0001204837224376267,
+      "loss": 1.7874,
+      "step": 1980
+    },
+    {
+      "epoch": 1.8586271305159934,
+      "grad_norm": 0.02644533256389969,
+      "learning_rate": 0.0001197268523105148,
+      "loss": 1.798,
+      "step": 1990
+    },
+    {
+      "epoch": 1.8679663787065142,
+      "grad_norm": 0.02999453651649614,
+      "learning_rate": 0.00011896880452149077,
+      "loss": 1.7957,
+      "step": 2000
+    },
+    {
+      "epoch": 1.877305626897035,
+      "grad_norm": 0.026905209700322272,
+      "learning_rate": 0.00011820962432480985,
+      "loss": 1.793,
+      "step": 2010
+    },
+    {
+      "epoch": 1.8866448750875553,
+      "grad_norm": 0.027263640323285022,
+      "learning_rate": 0.00011744935704233005,
+      "loss": 1.7974,
+      "step": 2020
+    },
+    {
+      "epoch": 1.895984123278076,
+      "grad_norm": 0.030479226063932337,
+      "learning_rate": 0.00011668804806080693,
+      "loss": 1.7898,
+      "step": 2030
+    },
+    {
+      "epoch": 1.9053233714685969,
+      "grad_norm": 0.030129902025534238,
+      "learning_rate": 0.00011592574282918369,
+      "loss": 1.7856,
+      "step": 2040
+    },
+    {
+      "epoch": 1.9146626196591174,
+      "grad_norm": 0.027884976674153635,
+      "learning_rate": 0.00011516248685587814,
+      "loss": 1.7858,
+      "step": 2050
+    },
+    {
+      "epoch": 1.924001867849638,
+      "grad_norm": 0.02925266011156687,
+      "learning_rate": 0.00011439832570606586,
+      "loss": 1.7876,
+      "step": 2060
+    },
+    {
+      "epoch": 1.9333411160401588,
+      "grad_norm": 0.028472914828616754,
+      "learning_rate": 0.00011363330499895997,
+      "loss": 1.7834,
+      "step": 2070
+    },
+    {
+      "epoch": 1.9426803642306796,
+      "grad_norm": 0.025877740032137875,
+      "learning_rate": 0.00011286747040508789,
+      "loss": 1.7955,
+      "step": 2080
+    },
+    {
+      "epoch": 1.9520196124212001,
+      "grad_norm": 0.02605295620697312,
+      "learning_rate": 0.0001121008676435648,
+      "loss": 1.7877,
+      "step": 2090
+    },
+    {
+      "epoch": 1.9613588606117207,
+      "grad_norm": 0.026887649929567867,
+      "learning_rate": 0.00011133354247936423,
+      "loss": 1.773,
+      "step": 2100
+    },
+    {
+      "epoch": 1.9706981088022415,
+      "grad_norm": 0.027982045915154026,
+      "learning_rate": 0.00011056554072058596,
+      "loss": 1.7762,
+      "step": 2110
+    },
+    {
+      "epoch": 1.980037356992762,
+      "grad_norm": 0.028077027059053006,
+      "learning_rate": 0.0001097969082157215,
+      "loss": 1.7963,
+      "step": 2120
+    },
+    {
+      "epoch": 1.9893766051832826,
+      "grad_norm": 0.027877078975954036,
+      "learning_rate": 0.00010902769085091686,
+      "loss": 1.7787,
+      "step": 2130
+    },
+    {
+      "epoch": 1.9987158533738034,
+      "grad_norm": 0.026120077388738373,
+      "learning_rate": 0.00010825793454723325,
+      "loss": 1.7842,
+      "step": 2140
+    },
+    {
+      "epoch": 2.0081718421667056,
+      "grad_norm": 0.044810283079268924,
+      "learning_rate": 0.00010748768525790569,
+      "loss": 1.6591,
+      "step": 2150
+    },
+    {
+      "epoch": 2.0175110903572264,
+      "grad_norm": 0.03431848280739808,
+      "learning_rate": 0.00010671698896559968,
+      "loss": 1.5599,
+      "step": 2160
+    },
+    {
+      "epoch": 2.0268503385477468,
+      "grad_norm": 0.04234332973849956,
+      "learning_rate": 0.00010594589167966606,
+      "loss": 1.5494,
+      "step": 2170
+    },
+    {
+      "epoch": 2.0361895867382676,
+      "grad_norm": 0.03260321438171042,
+      "learning_rate": 0.00010517443943339438,
+      "loss": 1.5473,
+      "step": 2180
+    },
+    {
+      "epoch": 2.0455288349287883,
+      "grad_norm": 0.034116901745609114,
+      "learning_rate": 0.00010440267828126478,
+      "loss": 1.5464,
+      "step": 2190
+    },
+    {
+      "epoch": 2.0548680831193087,
+      "grad_norm": 0.030992757239375807,
+      "learning_rate": 0.00010363065429619858,
+      "loss": 1.5514,
+      "step": 2200
+    },
+    {
+      "epoch": 2.0642073313098295,
+      "grad_norm": 0.03365516197786113,
+      "learning_rate": 0.0001028584135668077,
+      "loss": 1.5493,
+      "step": 2210
+    },
+    {
+      "epoch": 2.0735465795003503,
+      "grad_norm": 0.033293307482261586,
+      "learning_rate": 0.00010208600219464355,
+      "loss": 1.5426,
+      "step": 2220
+    },
+    {
+      "epoch": 2.082885827690871,
+      "grad_norm": 0.03653097737467338,
+      "learning_rate": 0.00010131346629144451,
+      "loss": 1.5471,
+      "step": 2230
+    },
+    {
+      "epoch": 2.0922250758813914,
+      "grad_norm": 0.03390291511697895,
+      "learning_rate": 0.0001005408519763833,
+      "loss": 1.5568,
+      "step": 2240
+    },
+    {
+      "epoch": 2.101564324071912,
+      "grad_norm": 0.03192694661852283,
+      "learning_rate": 9.976820537331374e-05,
+      "loss": 1.5452,
+      "step": 2250
+    },
+    {
+      "epoch": 2.110903572262433,
+      "grad_norm": 0.03561740193515691,
+      "learning_rate": 9.899557260801707e-05,
+      "loss": 1.546,
+      "step": 2260
+    },
+    {
+      "epoch": 2.1202428204529538,
+      "grad_norm": 0.029803732953068658,
+      "learning_rate": 9.822299980544862e-05,
+      "loss": 1.5533,
+      "step": 2270
+    },
+    {
+      "epoch": 2.129582068643474,
+      "grad_norm": 0.031232417271289125,
+      "learning_rate": 9.745053308698392e-05,
+      "loss": 1.5469,
+      "step": 2280
+    },
+    {
+      "epoch": 2.138921316833995,
+      "grad_norm": 0.032434793780181034,
+      "learning_rate": 9.667821856766548e-05,
+      "loss": 1.5514,
+      "step": 2290
+    },
+    {
+      "epoch": 2.1482605650245157,
+      "grad_norm": 0.03579370906405582,
+      "learning_rate": 9.590610235344972e-05,
+      "loss": 1.5577,
+      "step": 2300
+    },
+    {
+      "epoch": 2.157599813215036,
+      "grad_norm": 0.029662202478648328,
+      "learning_rate": 9.51342305384546e-05,
+      "loss": 1.5543,
+      "step": 2310
+    },
+    {
+      "epoch": 2.166939061405557,
+      "grad_norm": 0.03178715913934592,
+      "learning_rate": 9.436264920220781e-05,
+      "loss": 1.5579,
+      "step": 2320
+    },
+    {
+      "epoch": 2.1762783095960776,
+      "grad_norm": 0.03384008887051677,
+      "learning_rate": 9.359140440689601e-05,
+      "loss": 1.5595,
+      "step": 2330
+    },
+    {
+      "epoch": 2.1856175577865984,
+      "grad_norm": 0.03316450664408166,
+      "learning_rate": 9.282054219461475e-05,
+      "loss": 1.5556,
+      "step": 2340
+    },
+    {
+      "epoch": 2.1949568059771187,
+      "grad_norm": 0.032176305552558876,
+      "learning_rate": 9.205010858462007e-05,
+      "loss": 1.5638,
+      "step": 2350
+    },
+    {
+      "epoch": 2.2042960541676395,
+      "grad_norm": 0.031134335256756362,
+      "learning_rate": 9.128014957058109e-05,
+      "loss": 1.5629,
+      "step": 2360
+    },
+    {
+      "epoch": 2.2136353023581603,
+      "grad_norm": 0.032205851810441756,
+      "learning_rate": 9.051071111783436e-05,
+      "loss": 1.5613,
+      "step": 2370
+    },
+    {
+      "epoch": 2.2229745505486807,
+      "grad_norm": 0.029048245476020442,
+      "learning_rate": 8.974183916063968e-05,
+      "loss": 1.5594,
+      "step": 2380
+    },
+    {
+      "epoch": 2.2323137987392014,
+      "grad_norm": 0.03183166054573621,
+      "learning_rate": 8.897357959943795e-05,
+      "loss": 1.5606,
+      "step": 2390
+    },
+    {
+      "epoch": 2.2416530469297222,
+      "grad_norm": 0.032309922117136916,
+      "learning_rate": 8.820597829811109e-05,
+      "loss": 1.5524,
+      "step": 2400
+    },
+    {
+      "epoch": 2.250992295120243,
+      "grad_norm": 0.03598922231958808,
+      "learning_rate": 8.743908108124388e-05,
+      "loss": 1.5604,
+      "step": 2410
+    },
+    {
+      "epoch": 2.2603315433107634,
+      "grad_norm": 0.03046424735786346,
+      "learning_rate": 8.667293373138835e-05,
+      "loss": 1.5598,
+      "step": 2420
+    },
+    {
+      "epoch": 2.269670791501284,
+      "grad_norm": 0.030995453538377543,
+      "learning_rate": 8.59075819863307e-05,
+      "loss": 1.5652,
+      "step": 2430
+    },
+    {
+      "epoch": 2.279010039691805,
+      "grad_norm": 0.029309020623010097,
+      "learning_rate": 8.514307153636077e-05,
+      "loss": 1.5651,
+      "step": 2440
+    },
+    {
+      "epoch": 2.2883492878823253,
+      "grad_norm": 0.03158721106736763,
+      "learning_rate": 8.437944802154434e-05,
+      "loss": 1.5581,
+      "step": 2450
+    },
+    {
+      "epoch": 2.297688536072846,
+      "grad_norm": 0.03168229084049938,
+      "learning_rate": 8.361675702899871e-05,
+      "loss": 1.5671,
+      "step": 2460
+    },
+    {
+      "epoch": 2.307027784263367,
+      "grad_norm": 0.031335222148495136,
+      "learning_rate": 8.2855044090171e-05,
+      "loss": 1.5675,
+      "step": 2470
+    },
+    {
+      "epoch": 2.3163670324538876,
+      "grad_norm": 0.031073847325941303,
+      "learning_rate": 8.209435467811998e-05,
+      "loss": 1.5624,
+      "step": 2480
+    },
+    {
+      "epoch": 2.325706280644408,
+      "grad_norm": 0.030099100631045844,
+      "learning_rate": 8.133473420480161e-05,
+      "loss": 1.5606,
+      "step": 2490
+    },
+    {
+      "epoch": 2.3350455288349288,
+      "grad_norm": 0.034147632950361176,
+      "learning_rate": 8.057622801835788e-05,
+      "loss": 1.5703,
+      "step": 2500
+    },
+    {
+      "epoch": 2.3443847770254496,
+      "grad_norm": 0.03051580784550685,
+      "learning_rate": 7.981888140040955e-05,
+      "loss": 1.5731,
+      "step": 2510
+    },
+    {
+      "epoch": 2.3537240252159704,
+      "grad_norm": 0.03068917065832597,
+      "learning_rate": 7.9062739563353e-05,
+      "loss": 1.5723,
+      "step": 2520
+    },
+    {
+      "epoch": 2.3630632734064907,
+      "grad_norm": 0.02899547641705554,
+      "learning_rate": 7.830784764766118e-05,
+      "loss": 1.5691,
+      "step": 2530
+    },
+    {
+      "epoch": 2.3724025215970115,
+      "grad_norm": 0.030965383166701443,
+      "learning_rate": 7.755425071918858e-05,
+      "loss": 1.5627,
+      "step": 2540
+    },
+    {
+      "epoch": 2.3817417697875323,
+      "grad_norm": 0.03252440018336625,
+      "learning_rate": 7.680199376648108e-05,
+      "loss": 1.5536,
+      "step": 2550
+    },
+    {
+      "epoch": 2.3910810179780526,
+      "grad_norm": 0.031720485449340044,
+      "learning_rate": 7.605112169809008e-05,
+      "loss": 1.5617,
+      "step": 2560
+    },
+    {
+      "epoch": 2.4004202661685734,
+      "grad_norm": 0.031796658969132544,
+      "learning_rate": 7.530167933989161e-05,
+      "loss": 1.5595,
+      "step": 2570
+    },
+    {
+      "epoch": 2.409759514359094,
+      "grad_norm": 0.03218288097429844,
+      "learning_rate": 7.45537114324102e-05,
+      "loss": 1.5628,
+      "step": 2580
+    },
+    {
+      "epoch": 2.4190987625496145,
+      "grad_norm": 0.0305713183075559,
+      "learning_rate": 7.380726262814814e-05,
+      "loss": 1.5717,
+      "step": 2590
+    },
+    {
+      "epoch": 2.4284380107401353,
+      "grad_norm": 0.06879342166341705,
+      "learning_rate": 7.30623774889195e-05,
+      "loss": 1.5726,
+      "step": 2600
+    },
+    {
+      "epoch": 2.437777258930656,
+      "grad_norm": 0.04101428600237338,
+      "learning_rate": 7.231910048319011e-05,
+      "loss": 1.5679,
+      "step": 2610
+    },
+    {
+      "epoch": 2.447116507121177,
+      "grad_norm": 0.031060002638443395,
+      "learning_rate": 7.157747598342274e-05,
+      "loss": 1.562,
+      "step": 2620
+    },
+    {
+      "epoch": 2.4564557553116972,
+      "grad_norm": 0.032302829437386466,
+      "learning_rate": 7.083754826342816e-05,
+      "loss": 1.5767,
+      "step": 2630
+    },
+    {
+      "epoch": 2.465795003502218,
+      "grad_norm": 0.03111462744196413,
+      "learning_rate": 7.009936149572205e-05,
+      "loss": 1.5672,
+      "step": 2640
+    },
+    {
+      "epoch": 2.475134251692739,
+      "grad_norm": 0.031134083521743777,
+      "learning_rate": 6.936295974888807e-05,
+      "loss": 1.5665,
+      "step": 2650
+    },
+    {
+      "epoch": 2.4844734998832596,
+      "grad_norm": 0.030961556373721985,
+      "learning_rate": 6.862838698494693e-05,
+      "loss": 1.5608,
+      "step": 2660
+    },
+    {
+      "epoch": 2.49381274807378,
+      "grad_norm": 0.03168121700432082,
+      "learning_rate": 6.789568705673183e-05,
+      "loss": 1.566,
+      "step": 2670
+    },
+    {
+      "epoch": 2.5031519962643007,
+      "grad_norm": 0.030850372541726772,
+      "learning_rate": 6.716490370527081e-05,
+      "loss": 1.5651,
+      "step": 2680
+    },
+    {
+      "epoch": 2.5124912444548215,
+      "grad_norm": 0.03076635908430861,
+      "learning_rate": 6.643608055717519e-05,
+      "loss": 1.5596,
+      "step": 2690
+    },
+    {
+      "epoch": 2.521830492645342,
+      "grad_norm": 0.031897253741779714,
+      "learning_rate": 6.570926112203528e-05,
+      "loss": 1.5716,
+      "step": 2700
+    },
+    {
+      "epoch": 2.5311697408358627,
+      "grad_norm": 0.03085546721246857,
+      "learning_rate": 6.498448878982291e-05,
+      "loss": 1.5647,
+      "step": 2710
+    },
+    {
+      "epoch": 2.5405089890263834,
+      "grad_norm": 0.03127518794548787,
+      "learning_rate": 6.426180682830107e-05,
+      "loss": 1.5573,
+      "step": 2720
+    },
+    {
+      "epoch": 2.549848237216904,
+      "grad_norm": 0.03196649247686066,
+      "learning_rate": 6.354125838044098e-05,
+      "loss": 1.5597,
+      "step": 2730
+    },
+    {
+      "epoch": 2.5591874854074246,
+      "grad_norm": 0.030359755432035333,
+      "learning_rate": 6.282288646184638e-05,
+      "loss": 1.5625,
+      "step": 2740
+    },
+    {
+      "epoch": 2.5685267335979454,
+      "grad_norm": 0.03030640438940187,
+      "learning_rate": 6.210673395818571e-05,
+      "loss": 1.5717,
+      "step": 2750
+    },
+    {
+      "epoch": 2.577865981788466,
+      "grad_norm": 0.032197470232298186,
+      "learning_rate": 6.139284362263185e-05,
+      "loss": 1.5663,
+      "step": 2760
+    },
+    {
+      "epoch": 2.587205229978987,
+      "grad_norm": 0.030983733397891462,
+      "learning_rate": 6.0681258073309756e-05,
+      "loss": 1.5657,
+      "step": 2770
+    },
+    {
+      "epoch": 2.5965444781695073,
+      "grad_norm": 0.030427577702286164,
+      "learning_rate": 5.9972019790752385e-05,
+      "loss": 1.5708,
+      "step": 2780
+    },
+    {
+      "epoch": 2.605883726360028,
+      "grad_norm": 0.032761226318855745,
+      "learning_rate": 5.9265171115364495e-05,
+      "loss": 1.5641,
+      "step": 2790
+    },
+    {
+      "epoch": 2.615222974550549,
+      "grad_norm": 0.0317533648622182,
+      "learning_rate": 5.856075424489511e-05,
+      "loss": 1.5613,
+      "step": 2800
+    },
+    {
+      "epoch": 2.624562222741069,
+      "grad_norm": 0.03251887796623927,
+      "learning_rate": 5.785881123191834e-05,
+      "loss": 1.5644,
+      "step": 2810
+    },
+    {
+      "epoch": 2.63390147093159,
+      "grad_norm": 0.033109556138937665,
+      "learning_rate": 5.7159383981322866e-05,
+      "loss": 1.5613,
+      "step": 2820
+    },
+    {
+      "epoch": 2.643240719122111,
+      "grad_norm": 0.03190494841434546,
+      "learning_rate": 5.646251424781044e-05,
+      "loss": 1.5597,
+      "step": 2830
+    },
+    {
+      "epoch": 2.652579967312631,
+      "grad_norm": 0.03131425467271855,
+      "learning_rate": 5.576824363340293e-05,
+      "loss": 1.5644,
+      "step": 2840
+    },
+    {
+      "epoch": 2.661919215503152,
+      "grad_norm": 0.032580122114687284,
+      "learning_rate": 5.507661358495904e-05,
+      "loss": 1.5651,
+      "step": 2850
+    },
+    {
+      "epoch": 2.6712584636936727,
+      "grad_norm": 0.031118294785578867,
+      "learning_rate": 5.4387665391699814e-05,
+      "loss": 1.5595,
+      "step": 2860
+    },
+    {
+      "epoch": 2.680597711884193,
+      "grad_norm": 0.030454480461742643,
+      "learning_rate": 5.370144018274371e-05,
+      "loss": 1.5607,
+      "step": 2870
+    },
+    {
+      "epoch": 2.689936960074714,
+      "grad_norm": 0.030828671799046907,
+      "learning_rate": 5.301797892465148e-05,
+      "loss": 1.5587,
+      "step": 2880
+    },
+    {
+      "epoch": 2.6992762082652346,
+      "grad_norm": 0.030541588338347705,
+      "learning_rate": 5.2337322418980204e-05,
+      "loss": 1.5698,
+      "step": 2890
+    },
+    {
+      "epoch": 2.7086154564557554,
+      "grad_norm": 0.030634684731728472,
+      "learning_rate": 5.16595112998477e-05,
+      "loss": 1.5628,
+      "step": 2900
+    },
+    {
+      "epoch": 2.717954704646276,
+      "grad_norm": 0.03045168911848818,
+      "learning_rate": 5.098458603150691e-05,
+      "loss": 1.5544,
+      "step": 2910
+    },
+    {
+      "epoch": 2.7272939528367965,
+      "grad_norm": 0.029966073421542574,
+      "learning_rate": 5.0312586905929816e-05,
+      "loss": 1.557,
+      "step": 2920
+    },
+    {
+      "epoch": 2.7366332010273173,
+      "grad_norm": 0.031203173596988765,
+      "learning_rate": 4.964355404040232e-05,
+      "loss": 1.5571,
+      "step": 2930
+    },
+    {
+      "epoch": 2.745972449217838,
+      "grad_norm": 0.02933060802902336,
+      "learning_rate": 4.897752737512944e-05,
+      "loss": 1.5518,
+      "step": 2940
+    },
+    {
+      "epoch": 2.7553116974083585,
+      "grad_norm": 0.03039237943996916,
+      "learning_rate": 4.8314546670850594e-05,
+      "loss": 1.5682,
+      "step": 2950
+    },
+    {
+      "epoch": 2.7646509455988793,
+      "grad_norm": 0.03077504425768828,
+      "learning_rate": 4.765465150646633e-05,
+      "loss": 1.5618,
+      "step": 2960
+    },
+    {
+      "epoch": 2.7739901937894,
+      "grad_norm": 0.02987361489256037,
+      "learning_rate": 4.699788127667517e-05,
+      "loss": 1.5657,
+      "step": 2970
+    },
+    {
+      "epoch": 2.7833294419799204,
+      "grad_norm": 0.030823634404763923,
+      "learning_rate": 4.634427518962209e-05,
+      "loss": 1.5611,
+      "step": 2980
+    },
+    {
+      "epoch": 2.792668690170441,
+      "grad_norm": 0.031136974955910973,
+      "learning_rate": 4.569387226455776e-05,
+      "loss": 1.558,
+      "step": 2990
+    },
+    {
+      "epoch": 2.802007938360962,
+      "grad_norm": 0.032345263568839946,
+      "learning_rate": 4.5046711329508997e-05,
+      "loss": 1.5567,
+      "step": 3000
+    },
+    {
+      "epoch": 2.8113471865514827,
+      "grad_norm": 0.03075590840408306,
+      "learning_rate": 4.440283101896112e-05,
+      "loss": 1.5509,
+      "step": 3010
+    },
+    {
+      "epoch": 2.8206864347420035,
+      "grad_norm": 0.03212380955577142,
+      "learning_rate": 4.376226977155118e-05,
+      "loss": 1.5549,
+      "step": 3020
+    },
+    {
+      "epoch": 2.830025682932524,
+      "grad_norm": 0.030448994102361984,
+      "learning_rate": 4.3125065827773535e-05,
+      "loss": 1.5582,
+      "step": 3030
+    },
+    {
+      "epoch": 2.8393649311230447,
+      "grad_norm": 0.02952309346631166,
+      "learning_rate": 4.249125722769679e-05,
+      "loss": 1.5621,
+      "step": 3040
+    },
+    {
+      "epoch": 2.8487041793135655,
+      "grad_norm": 0.029801760584304334,
+      "learning_rate": 4.18608818086928e-05,
+      "loss": 1.5566,
+      "step": 3050
+    },
+    {
+      "epoch": 2.858043427504086,
+      "grad_norm": 0.029682850572275397,
+      "learning_rate": 4.12339772031781e-05,
+      "loss": 1.563,
+      "step": 3060
+    },
+    {
+      "epoch": 2.8673826756946066,
+      "grad_norm": 0.0295800536252474,
+      "learning_rate": 4.061058083636702e-05,
+      "loss": 1.5478,
+      "step": 3070
+    },
+    {
+      "epoch": 2.8767219238851274,
+      "grad_norm": 0.029683305961870896,
+      "learning_rate": 3.999072992403756e-05,
+      "loss": 1.5628,
+      "step": 3080
+    },
+    {
+      "epoch": 2.8860611720756477,
+      "grad_norm": 0.029644699672258072,
+      "learning_rate": 3.93744614703098e-05,
+      "loss": 1.5541,
+      "step": 3090
+    },
+    {
+      "epoch": 2.8954004202661685,
+      "grad_norm": 0.030518095362273494,
+      "learning_rate": 3.876181226543668e-05,
+      "loss": 1.5702,
+      "step": 3100
+    },
+    {
+      "epoch": 2.9047396684566893,
+      "grad_norm": 0.030536745251161155,
+      "learning_rate": 3.81528188836076e-05,
+      "loss": 1.5551,
+      "step": 3110
+    },
+    {
+      "epoch": 2.9140789166472096,
+      "grad_norm": 0.030847847626104093,
+      "learning_rate": 3.7547517680765244e-05,
+      "loss": 1.5595,
+      "step": 3120
+    },
+    {
+      "epoch": 2.9234181648377304,
+      "grad_norm": 0.029492648716320746,
+      "learning_rate": 3.6945944792434906e-05,
+      "loss": 1.5536,
+      "step": 3130
+    },
+    {
+      "epoch": 2.932757413028251,
+      "grad_norm": 0.03032565633382453,
+      "learning_rate": 3.634813613156753e-05,
+      "loss": 1.5545,
+      "step": 3140
+    },
+    {
+      "epoch": 2.942096661218772,
+      "grad_norm": 0.031152780923178743,
+      "learning_rate": 3.5754127386395496e-05,
+      "loss": 1.5553,
+      "step": 3150
+    },
+    {
+      "epoch": 2.951435909409293,
+      "grad_norm": 0.03107457614016718,
+      "learning_rate": 3.5163954018302313e-05,
+      "loss": 1.5612,
+      "step": 3160
+    },
+    {
+      "epoch": 2.960775157599813,
+      "grad_norm": 0.03104057970445527,
+      "learning_rate": 3.4577651259705545e-05,
+      "loss": 1.5461,
+      "step": 3170
+    },
+    {
+      "epoch": 2.970114405790334,
+      "grad_norm": 0.03135460216536421,
+      "learning_rate": 3.399525411195339e-05,
+      "loss": 1.556,
+      "step": 3180
+    },
+    {
+      "epoch": 2.9794536539808547,
+      "grad_norm": 0.03023549504443143,
+      "learning_rate": 3.3416797343235375e-05,
+      "loss": 1.5513,
+      "step": 3190
+    },
+    {
+      "epoch": 2.988792902171375,
+      "grad_norm": 0.030100263338637512,
+      "learning_rate": 3.284231548650649e-05,
+      "loss": 1.5646,
+      "step": 3200
+    },
+    {
+      "epoch": 2.998132150361896,
+      "grad_norm": 0.03034217414728692,
+      "learning_rate": 3.227184283742591e-05,
+      "loss": 1.5521,
+      "step": 3210
+    },
+    {
+      "epoch": 3.007588139154798,
+      "grad_norm": 0.06632657839957506,
+      "learning_rate": 3.17054134523093e-05,
+      "loss": 1.3861,
+      "step": 3220
+    },
+    {
+      "epoch": 3.016927387345319,
+      "grad_norm": 0.04494797332190053,
+      "learning_rate": 3.114306114609594e-05,
+      "loss": 1.2688,
+      "step": 3230
+    },
+    {
+      "epoch": 3.026266635535839,
+      "grad_norm": 0.04243454415323923,
+      "learning_rate": 3.058481949032998e-05,
+      "loss": 1.2568,
+      "step": 3240
+    },
+    {
+      "epoch": 3.03560588372636,
+      "grad_norm": 0.037189416438265135,
+      "learning_rate": 3.003072181115615e-05,
+      "loss": 1.2674,
+      "step": 3250
+    },
+    {
+      "epoch": 3.0449451319168808,
+      "grad_norm": 0.036342648540388975,
+      "learning_rate": 2.9480801187330308e-05,
+      "loss": 1.2553,
+      "step": 3260
+    },
+    {
+      "epoch": 3.0542843801074016,
+      "grad_norm": 0.03758108482902846,
+      "learning_rate": 2.8935090448244838e-05,
+      "loss": 1.2587,
+      "step": 3270
+    },
+    {
+      "epoch": 3.063623628297922,
+      "grad_norm": 0.03779156244551835,
+      "learning_rate": 2.8393622171968494e-05,
+      "loss": 1.2615,
+      "step": 3280
+    },
+    {
+      "epoch": 3.0729628764884427,
+      "grad_norm": 0.0354713302142897,
+      "learning_rate": 2.7856428683301827e-05,
+      "loss": 1.2552,
+      "step": 3290
+    },
+    {
+      "epoch": 3.0823021246789635,
+      "grad_norm": 0.03861733806219831,
+      "learning_rate": 2.7323542051847305e-05,
+      "loss": 1.2591,
+      "step": 3300
+    },
+    {
+      "epoch": 3.091641372869484,
+      "grad_norm": 0.03927047617762108,
+      "learning_rate": 2.6794994090094748e-05,
+      "loss": 1.2568,
+      "step": 3310
+    },
+    {
+      "epoch": 3.1009806210600046,
+      "grad_norm": 0.03904084831655504,
+      "learning_rate": 2.6270816351522352e-05,
+      "loss": 1.2548,
+      "step": 3320
+    },
+    {
+      "epoch": 3.1103198692505254,
+      "grad_norm": 0.037088038012932166,
+      "learning_rate": 2.57510401287128e-05,
+      "loss": 1.2535,
+      "step": 3330
+    },
+    {
+      "epoch": 3.119659117441046,
+      "grad_norm": 0.039689764947511294,
+      "learning_rate": 2.5235696451485334e-05,
+      "loss": 1.2527,
+      "step": 3340
+    },
+    {
+      "epoch": 3.1289983656315665,
+      "grad_norm": 0.03823754136616741,
+      "learning_rate": 2.4724816085043122e-05,
+      "loss": 1.2614,
+      "step": 3350
+    },
+    {
+      "epoch": 3.1383376138220873,
+      "grad_norm": 0.03869724476981111,
+      "learning_rate": 2.421842952813683e-05,
+      "loss": 1.2539,
+      "step": 3360
+    },
+    {
+      "epoch": 3.147676862012608,
+      "grad_norm": 0.03803608771547874,
+      "learning_rate": 2.371656701124376e-05,
+      "loss": 1.2491,
+      "step": 3370
+    },
+    {
+      "epoch": 3.1570161102031284,
+      "grad_norm": 0.03931544384220963,
+      "learning_rate": 2.321925849476314e-05,
+      "loss": 1.2505,
+      "step": 3380
+    },
+    {
+      "epoch": 3.1663553583936492,
+      "grad_norm": 0.039905322735836295,
+      "learning_rate": 2.2726533667227623e-05,
+      "loss": 1.2585,
+      "step": 3390
+    },
+    {
+      "epoch": 3.17569460658417,
+      "grad_norm": 0.03812720334866993,
+      "learning_rate": 2.223842194353084e-05,
+      "loss": 1.2589,
+      "step": 3400
+    },
+    {
+      "epoch": 3.185033854774691,
+      "grad_norm": 0.03970128613042995,
+      "learning_rate": 2.1754952463171385e-05,
+      "loss": 1.2547,
+      "step": 3410
+    },
+    {
+      "epoch": 3.194373102965211,
+      "grad_norm": 0.03968022448776454,
+      "learning_rate": 2.1276154088513322e-05,
+      "loss": 1.256,
+      "step": 3420
+    },
+    {
+      "epoch": 3.203712351155732,
+      "grad_norm": 0.038047431666726306,
+      "learning_rate": 2.080205540306307e-05,
+      "loss": 1.2618,
+      "step": 3430
+    },
+    {
+      "epoch": 3.2130515993462527,
+      "grad_norm": 0.03796892071652476,
+      "learning_rate": 2.0332684709763037e-05,
+      "loss": 1.258,
+      "step": 3440
+    },
+    {
+      "epoch": 3.222390847536773,
+      "grad_norm": 0.03743002539050354,
+      "learning_rate": 1.986807002930192e-05,
+      "loss": 1.2625,
+      "step": 3450
+    },
+    {
+      "epoch": 3.231730095727294,
+      "grad_norm": 0.03845080735099829,
+      "learning_rate": 1.940823909844198e-05,
+      "loss": 1.2583,
+      "step": 3460
+    },
+    {
+      "epoch": 3.2410693439178146,
+      "grad_norm": 0.0384612122659626,
+      "learning_rate": 1.895321936836324e-05,
+      "loss": 1.2504,
+      "step": 3470
+    },
+    {
+      "epoch": 3.2504085921083354,
+      "grad_norm": 0.03770473444960908,
+      "learning_rate": 1.8503038003024564e-05,
+      "loss": 1.2691,
+      "step": 3480
+    },
+    {
+      "epoch": 3.259747840298856,
+      "grad_norm": 0.0368880551504251,
+      "learning_rate": 1.8057721877542135e-05,
+      "loss": 1.2562,
+      "step": 3490
+    },
+    {
+      "epoch": 3.2690870884893766,
+      "grad_norm": 0.035849085577933516,
+      "learning_rate": 1.7617297576585024e-05,
+      "loss": 1.2557,
+      "step": 3500
+    },
+    {
+      "epoch": 3.2784263366798974,
+      "grad_norm": 0.03880537892593414,
+      "learning_rate": 1.7181791392788016e-05,
+      "loss": 1.2606,
+      "step": 3510
+    },
+    {
+      "epoch": 3.287765584870418,
+      "grad_norm": 0.036791883288228966,
+      "learning_rate": 1.6751229325182195e-05,
+      "loss": 1.2633,
+      "step": 3520
+    },
+    {
+      "epoch": 3.2971048330609385,
+      "grad_norm": 0.03888037048657713,
+      "learning_rate": 1.6325637077642597e-05,
+      "loss": 1.2517,
+      "step": 3530
+    },
+    {
+      "epoch": 3.3064440812514593,
+      "grad_norm": 0.03624411339308035,
+      "learning_rate": 1.590504005735396e-05,
+      "loss": 1.2601,
+      "step": 3540
+    },
+    {
+      "epoch": 3.31578332944198,
+      "grad_norm": 0.037212088295175624,
+      "learning_rate": 1.5489463373293766e-05,
+      "loss": 1.2572,
+      "step": 3550
+    },
+    {
+      "epoch": 3.3251225776325004,
+      "grad_norm": 0.037139053328387565,
+      "learning_rate": 1.5078931834733413e-05,
+      "loss": 1.2587,
+      "step": 3560
+    },
+    {
+      "epoch": 3.334461825823021,
+      "grad_norm": 0.038853212387254546,
+      "learning_rate": 1.467346994975708e-05,
+      "loss": 1.2598,
+      "step": 3570
+    },
+    {
+      "epoch": 3.343801074013542,
+      "grad_norm": 0.038068637710771305,
+      "learning_rate": 1.4273101923798593e-05,
+      "loss": 1.2511,
+      "step": 3580
+    },
+    {
+      "epoch": 3.3531403222040623,
+      "grad_norm": 0.03812184703208946,
+      "learning_rate": 1.3877851658196473e-05,
+      "loss": 1.264,
+      "step": 3590
+    },
+    {
+      "epoch": 3.362479570394583,
+      "grad_norm": 0.03544270078567587,
+      "learning_rate": 1.3487742748767052e-05,
+      "loss": 1.2566,
+      "step": 3600
+    },
+    {
+      "epoch": 3.371818818585104,
+      "grad_norm": 0.03729183894507396,
+      "learning_rate": 1.3102798484395761e-05,
+      "loss": 1.2554,
+      "step": 3610
+    },
+    {
+      "epoch": 3.3811580667756247,
+      "grad_norm": 0.03769688061241049,
+      "learning_rate": 1.2723041845646999e-05,
+      "loss": 1.2424,
+      "step": 3620
+    },
+    {
+      "epoch": 3.390497314966145,
+      "grad_norm": 0.03751980172858499,
+      "learning_rate": 1.2348495503391999e-05,
+      "loss": 1.2601,
+      "step": 3630
+    },
+    {
+      "epoch": 3.399836563156666,
+      "grad_norm": 0.03831727516561825,
+      "learning_rate": 1.1979181817455543e-05,
+      "loss": 1.2545,
+      "step": 3640
+    },
+    {
+      "epoch": 3.4091758113471866,
+      "grad_norm": 0.03929713475600568,
+      "learning_rate": 1.1615122835281156e-05,
+      "loss": 1.2608,
+      "step": 3650
+    },
+    {
+      "epoch": 3.4185150595377074,
+      "grad_norm": 0.04026553291819152,
+      "learning_rate": 1.1256340290614787e-05,
+      "loss": 1.2565,
+      "step": 3660
+    },
+    {
+      "epoch": 3.4278543077282277,
+      "grad_norm": 0.03669233459600138,
+      "learning_rate": 1.0902855602207451e-05,
+      "loss": 1.2512,
+      "step": 3670
+    },
+    {
+      "epoch": 3.4371935559187485,
+      "grad_norm": 0.036514113026692024,
+      "learning_rate": 1.0554689872536515e-05,
+      "loss": 1.2522,
+      "step": 3680
+    },
+    {
+      "epoch": 3.4465328041092693,
+      "grad_norm": 0.040574229602467475,
+      "learning_rate": 1.0211863886545859e-05,
+      "loss": 1.2653,
+      "step": 3690
+    },
+    {
+      "epoch": 3.4558720522997897,
+      "grad_norm": 0.037888410230506514,
+      "learning_rate": 9.87439811040518e-06,
+      "loss": 1.2553,
+      "step": 3700
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 4280,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 100,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 2.995895146589769e+20,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

l2-13b-ga/checkpoint-3700/zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

l2-13b-ga/checkpoint-4280/config.json

@@ -0,0 +1,30 @@
+{
+  "_name_or_path": "meta-llama/Llama-2-13b-hf",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 5120,
+  "initializer_range": 0.02,
+  "intermediate_size": 13824,
+  "max_position_embeddings": 4096,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 40,
+  "num_hidden_layers": 40,
+  "num_key_value_heads": 40,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.46.3",
+  "use_cache": true,
+  "vocab_size": 35483
+}

l2-13b-ga/checkpoint-4280/generation_config.json

@@ -0,0 +1,10 @@
+{
+  "bos_token_id": 1,
+  "do_sample": true,
+  "eos_token_id": 2,
+  "max_length": 4096,
+  "pad_token_id": 0,
+  "temperature": 0.6,
+  "top_p": 0.9,
+  "transformers_version": "4.46.3"
+}

l2-13b-ga/checkpoint-4280/latest

@@ -0,0 +1 @@
+global_step4280

l2-13b-ga/checkpoint-4280/model.safetensors.index.json

@@ -0,0 +1,370 @@
+{
+  "metadata": {
+    "total_size": 26103060480
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00006-of-00006.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.15.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.30.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.31.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.input_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.down_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.gate_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.up_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.post_attention_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.39.input_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.down_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.gate_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.up_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.post_attention_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.k_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.o_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.q_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.v_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.norm.weight": "model-00006-of-00006.safetensors"
+  }
+}

l2-13b-ga/checkpoint-4280/special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

l2-13b-ga/checkpoint-4280/tokenizer_config.json

@@ -0,0 +1,42 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": true,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": true,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

l2-13b-ga/checkpoint-4280/trainer_state.json

@@ -0,0 +1,3036 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 3.9975484473499883,
+  "eval_steps": 500,
+  "global_step": 4280,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0009339248190520663,
+      "grad_norm": 6.638877692627699,
+      "learning_rate": 9.345794392523364e-07,
+      "loss": 9.2917,
+      "step": 1
+    },
+    {
+      "epoch": 0.009339248190520663,
+      "grad_norm": 1.1560921335705272,
+      "learning_rate": 9.345794392523365e-06,
+      "loss": 9.0876,
+      "step": 10
+    },
+    {
+      "epoch": 0.018678496381041326,
+      "grad_norm": 0.8415132296956432,
+      "learning_rate": 1.869158878504673e-05,
+      "loss": 8.2164,
+      "step": 20
+    },
+    {
+      "epoch": 0.02801774457156199,
+      "grad_norm": 0.45381630992958155,
+      "learning_rate": 2.8037383177570094e-05,
+      "loss": 7.5184,
+      "step": 30
+    },
+    {
+      "epoch": 0.03735699276208265,
+      "grad_norm": 0.8400636107958425,
+      "learning_rate": 3.738317757009346e-05,
+      "loss": 6.6507,
+      "step": 40
+    },
+    {
+      "epoch": 0.046696240952603316,
+      "grad_norm": 0.557696240829066,
+      "learning_rate": 4.672897196261683e-05,
+      "loss": 5.8909,
+      "step": 50
+    },
+    {
+      "epoch": 0.05603548914312398,
+      "grad_norm": 0.3971996057467842,
+      "learning_rate": 5.607476635514019e-05,
+      "loss": 5.4127,
+      "step": 60
+    },
+    {
+      "epoch": 0.06537473733364464,
+      "grad_norm": 0.2932710540265688,
+      "learning_rate": 6.542056074766355e-05,
+      "loss": 5.0106,
+      "step": 70
+    },
+    {
+      "epoch": 0.0747139855241653,
+      "grad_norm": 0.3682690443551033,
+      "learning_rate": 7.476635514018692e-05,
+      "loss": 4.6042,
+      "step": 80
+    },
+    {
+      "epoch": 0.08405323371468597,
+      "grad_norm": 0.3132971920011515,
+      "learning_rate": 8.411214953271028e-05,
+      "loss": 4.2031,
+      "step": 90
+    },
+    {
+      "epoch": 0.09339248190520663,
+      "grad_norm": 0.6731868159213446,
+      "learning_rate": 9.345794392523365e-05,
+      "loss": 3.9423,
+      "step": 100
+    },
+    {
+      "epoch": 0.1027317300957273,
+      "grad_norm": 0.27848867836763197,
+      "learning_rate": 0.000102803738317757,
+      "loss": 3.7157,
+      "step": 110
+    },
+    {
+      "epoch": 0.11207097828624796,
+      "grad_norm": 0.24642109032991807,
+      "learning_rate": 0.00011214953271028037,
+      "loss": 3.4516,
+      "step": 120
+    },
+    {
+      "epoch": 0.12141022647676862,
+      "grad_norm": 0.25717384664029797,
+      "learning_rate": 0.00012149532710280373,
+      "loss": 3.2167,
+      "step": 130
+    },
+    {
+      "epoch": 0.13074947466728928,
+      "grad_norm": 0.20912922668565637,
+      "learning_rate": 0.0001308411214953271,
+      "loss": 3.0237,
+      "step": 140
+    },
+    {
+      "epoch": 0.14008872285780993,
+      "grad_norm": 0.15805888388706113,
+      "learning_rate": 0.00014018691588785047,
+      "loss": 2.8529,
+      "step": 150
+    },
+    {
+      "epoch": 0.1494279710483306,
+      "grad_norm": 0.23370349497479534,
+      "learning_rate": 0.00014953271028037384,
+      "loss": 2.7078,
+      "step": 160
+    },
+    {
+      "epoch": 0.15876721923885126,
+      "grad_norm": 0.1802138633012483,
+      "learning_rate": 0.0001588785046728972,
+      "loss": 2.6115,
+      "step": 170
+    },
+    {
+      "epoch": 0.16810646742937194,
+      "grad_norm": 0.13354347610039718,
+      "learning_rate": 0.00016822429906542056,
+      "loss": 2.5309,
+      "step": 180
+    },
+    {
+      "epoch": 0.17744571561989259,
+      "grad_norm": 0.09414865188086892,
+      "learning_rate": 0.00017757009345794393,
+      "loss": 2.4452,
+      "step": 190
+    },
+    {
+      "epoch": 0.18678496381041326,
+      "grad_norm": 0.08333601554768896,
+      "learning_rate": 0.0001869158878504673,
+      "loss": 2.3832,
+      "step": 200
+    },
+    {
+      "epoch": 0.1961242120009339,
+      "grad_norm": 0.15926414699806835,
+      "learning_rate": 0.00019626168224299065,
+      "loss": 2.3492,
+      "step": 210
+    },
+    {
+      "epoch": 0.2054634601914546,
+      "grad_norm": 0.09492820761057012,
+      "learning_rate": 0.0001999989254250208,
+      "loss": 2.323,
+      "step": 220
+    },
+    {
+      "epoch": 0.21480270838197524,
+      "grad_norm": 0.0801349259356147,
+      "learning_rate": 0.00019999235866155886,
+      "loss": 2.2731,
+      "step": 230
+    },
+    {
+      "epoch": 0.22414195657249592,
+      "grad_norm": 0.12210960524693895,
+      "learning_rate": 0.00019997982251228469,
+      "loss": 2.2433,
+      "step": 240
+    },
+    {
+      "epoch": 0.23348120476301656,
+      "grad_norm": 3.14289498732125,
+      "learning_rate": 0.00019996131772558666,
+      "loss": 3.2769,
+      "step": 250
+    },
+    {
+      "epoch": 0.24282045295353724,
+      "grad_norm": 1.632940983166179,
+      "learning_rate": 0.00019993684540617132,
+      "loss": 4.9343,
+      "step": 260
+    },
+    {
+      "epoch": 0.2521597011440579,
+      "grad_norm": 3.4831252230225416,
+      "learning_rate": 0.00019990640701499736,
+      "loss": 4.2768,
+      "step": 270
+    },
+    {
+      "epoch": 0.26149894933457857,
+      "grad_norm": 1.6069045920523788,
+      "learning_rate": 0.00019987000436918874,
+      "loss": 5.9581,
+      "step": 280
+    },
+    {
+      "epoch": 0.27083819752509924,
+      "grad_norm": 0.2220907936615993,
+      "learning_rate": 0.00019982763964192585,
+      "loss": 3.8228,
+      "step": 290
+    },
+    {
+      "epoch": 0.28017744571561987,
+      "grad_norm": 0.24737284913291765,
+      "learning_rate": 0.00019977931536231596,
+      "loss": 3.1413,
+      "step": 300
+    },
+    {
+      "epoch": 0.28951669390614054,
+      "grad_norm": 4.010404518241152,
+      "learning_rate": 0.00019972503441524224,
+      "loss": 2.8432,
+      "step": 310
+    },
+    {
+      "epoch": 0.2988559420966612,
+      "grad_norm": 0.1515583580811596,
+      "learning_rate": 0.00019966480004119142,
+      "loss": 2.7859,
+      "step": 320
+    },
+    {
+      "epoch": 0.3081951902871819,
+      "grad_norm": 0.11259395750650594,
+      "learning_rate": 0.00019959861583606045,
+      "loss": 2.5821,
+      "step": 330
+    },
+    {
+      "epoch": 0.3175344384777025,
+      "grad_norm": 0.22514797814956813,
+      "learning_rate": 0.00019952648575094183,
+      "loss": 2.4517,
+      "step": 340
+    },
+    {
+      "epoch": 0.3268736866682232,
+      "grad_norm": 0.08040136172033542,
+      "learning_rate": 0.00019944841409188767,
+      "loss": 2.3794,
+      "step": 350
+    },
+    {
+      "epoch": 0.3362129348587439,
+      "grad_norm": 0.054758073593565354,
+      "learning_rate": 0.00019936440551965263,
+      "loss": 2.3232,
+      "step": 360
+    },
+    {
+      "epoch": 0.34555218304926455,
+      "grad_norm": 0.06742998909645591,
+      "learning_rate": 0.00019927446504941577,
+      "loss": 2.2776,
+      "step": 370
+    },
+    {
+      "epoch": 0.35489143123978517,
+      "grad_norm": 0.048780907584876736,
+      "learning_rate": 0.00019917859805048096,
+      "loss": 2.2376,
+      "step": 380
+    },
+    {
+      "epoch": 0.36423067943030585,
+      "grad_norm": 0.0475325963052214,
+      "learning_rate": 0.00019907681024595663,
+      "loss": 2.2191,
+      "step": 390
+    },
+    {
+      "epoch": 0.3735699276208265,
+      "grad_norm": 0.054089563211590065,
+      "learning_rate": 0.00019896910771241387,
+      "loss": 2.1961,
+      "step": 400
+    },
+    {
+      "epoch": 0.3829091758113472,
+      "grad_norm": 0.21798406131864823,
+      "learning_rate": 0.00019885549687952372,
+      "loss": 2.2078,
+      "step": 410
+    },
+    {
+      "epoch": 0.3922484240018678,
+      "grad_norm": 0.8673185709111124,
+      "learning_rate": 0.00019873598452967338,
+      "loss": 2.3731,
+      "step": 420
+    },
+    {
+      "epoch": 0.4015876721923885,
+      "grad_norm": 0.22424350669971718,
+      "learning_rate": 0.0001986105777975613,
+      "loss": 2.6195,
+      "step": 430
+    },
+    {
+      "epoch": 0.4109269203829092,
+      "grad_norm": 0.307418135168262,
+      "learning_rate": 0.00019847928416977126,
+      "loss": 2.3624,
+      "step": 440
+    },
+    {
+      "epoch": 0.42026616857342985,
+      "grad_norm": 0.07944722668080402,
+      "learning_rate": 0.00019834211148432536,
+      "loss": 2.2799,
+      "step": 450
+    },
+    {
+      "epoch": 0.4296054167639505,
+      "grad_norm": 0.18146933758664588,
+      "learning_rate": 0.00019819906793021614,
+      "loss": 2.2177,
+      "step": 460
+    },
+    {
+      "epoch": 0.43894466495447115,
+      "grad_norm": 0.07035825837333018,
+      "learning_rate": 0.0001980501620469178,
+      "loss": 2.1767,
+      "step": 470
+    },
+    {
+      "epoch": 0.44828391314499183,
+      "grad_norm": 0.04596186944454228,
+      "learning_rate": 0.0001978954027238763,
+      "loss": 2.1598,
+      "step": 480
+    },
+    {
+      "epoch": 0.4576231613355125,
+      "grad_norm": 0.041342347745088055,
+      "learning_rate": 0.0001977347991999786,
+      "loss": 2.131,
+      "step": 490
+    },
+    {
+      "epoch": 0.46696240952603313,
+      "grad_norm": 0.04172063219841485,
+      "learning_rate": 0.00019756836106300137,
+      "loss": 2.1231,
+      "step": 500
+    },
+    {
+      "epoch": 0.4763016577165538,
+      "grad_norm": 0.03373646457711144,
+      "learning_rate": 0.00019739609824903843,
+      "loss": 2.1146,
+      "step": 510
+    },
+    {
+      "epoch": 0.4856409059070745,
+      "grad_norm": 0.03736871030676605,
+      "learning_rate": 0.00019721802104190748,
+      "loss": 2.1003,
+      "step": 520
+    },
+    {
+      "epoch": 0.49498015409759516,
+      "grad_norm": 0.033931028038211034,
+      "learning_rate": 0.00019703414007253645,
+      "loss": 2.0983,
+      "step": 530
+    },
+    {
+      "epoch": 0.5043194022881158,
+      "grad_norm": 0.03790055446070549,
+      "learning_rate": 0.00019684446631832868,
+      "loss": 2.092,
+      "step": 540
+    },
+    {
+      "epoch": 0.5136586504786365,
+      "grad_norm": 0.030956192803893078,
+      "learning_rate": 0.00019664901110250758,
+      "loss": 2.0807,
+      "step": 550
+    },
+    {
+      "epoch": 0.5229978986691571,
+      "grad_norm": 0.03542530209935129,
+      "learning_rate": 0.00019644778609344068,
+      "loss": 2.0773,
+      "step": 560
+    },
+    {
+      "epoch": 0.5323371468596778,
+      "grad_norm": 0.040947757568902336,
+      "learning_rate": 0.00019624080330394306,
+      "loss": 2.0649,
+      "step": 570
+    },
+    {
+      "epoch": 0.5416763950501985,
+      "grad_norm": 0.034273415973688146,
+      "learning_rate": 0.00019602807509056018,
+      "loss": 2.0479,
+      "step": 580
+    },
+    {
+      "epoch": 0.5510156432407192,
+      "grad_norm": 0.031427481498873144,
+      "learning_rate": 0.00019580961415283028,
+      "loss": 2.0563,
+      "step": 590
+    },
+    {
+      "epoch": 0.5603548914312397,
+      "grad_norm": 0.03141549752041532,
+      "learning_rate": 0.00019558543353252611,
+      "loss": 2.0503,
+      "step": 600
+    },
+    {
+      "epoch": 0.5696941396217604,
+      "grad_norm": 0.033012392726428204,
+      "learning_rate": 0.00019535554661287652,
+      "loss": 2.0389,
+      "step": 610
+    },
+    {
+      "epoch": 0.5790333878122811,
+      "grad_norm": 0.02913261992661444,
+      "learning_rate": 0.0001951199671177673,
+      "loss": 2.036,
+      "step": 620
+    },
+    {
+      "epoch": 0.5883726360028018,
+      "grad_norm": 0.030543903708435332,
+      "learning_rate": 0.00019487870911092214,
+      "loss": 2.0326,
+      "step": 630
+    },
+    {
+      "epoch": 0.5977118841933224,
+      "grad_norm": 0.03215005545393897,
+      "learning_rate": 0.00019463178699506277,
+      "loss": 2.0231,
+      "step": 640
+    },
+    {
+      "epoch": 0.6070511323838431,
+      "grad_norm": 0.03823630791937631,
+      "learning_rate": 0.00019437921551104933,
+      "loss": 2.0293,
+      "step": 650
+    },
+    {
+      "epoch": 0.6163903805743638,
+      "grad_norm": 0.03200103149471209,
+      "learning_rate": 0.00019412100973700038,
+      "loss": 2.017,
+      "step": 660
+    },
+    {
+      "epoch": 0.6257296287648845,
+      "grad_norm": 0.03841804153577787,
+      "learning_rate": 0.00019385718508739262,
+      "loss": 2.0135,
+      "step": 670
+    },
+    {
+      "epoch": 0.635068876955405,
+      "grad_norm": 0.03052396655271533,
+      "learning_rate": 0.0001935877573121407,
+      "loss": 2.0237,
+      "step": 680
+    },
+    {
+      "epoch": 0.6444081251459257,
+      "grad_norm": 0.033817837533771815,
+      "learning_rate": 0.00019331274249565717,
+      "loss": 2.0069,
+      "step": 690
+    },
+    {
+      "epoch": 0.6537473733364464,
+      "grad_norm": 0.028286999650643876,
+      "learning_rate": 0.00019303215705589194,
+      "loss": 2.0112,
+      "step": 700
+    },
+    {
+      "epoch": 0.6630866215269671,
+      "grad_norm": 0.030693948626257357,
+      "learning_rate": 0.00019274601774335243,
+      "loss": 2.0,
+      "step": 710
+    },
+    {
+      "epoch": 0.6724258697174877,
+      "grad_norm": 0.029984615635953022,
+      "learning_rate": 0.0001924543416401035,
+      "loss": 2.0028,
+      "step": 720
+    },
+    {
+      "epoch": 0.6817651179080084,
+      "grad_norm": 0.02915985673921391,
+      "learning_rate": 0.00019215714615874755,
+      "loss": 2.0031,
+      "step": 730
+    },
+    {
+      "epoch": 0.6911043660985291,
+      "grad_norm": 0.028305399777245336,
+      "learning_rate": 0.00019185444904138528,
+      "loss": 1.9924,
+      "step": 740
+    },
+    {
+      "epoch": 0.7004436142890498,
+      "grad_norm": 0.036720505429756495,
+      "learning_rate": 0.00019154626835855628,
+      "loss": 1.9981,
+      "step": 750
+    },
+    {
+      "epoch": 0.7097828624795703,
+      "grad_norm": 0.0287124048917296,
+      "learning_rate": 0.00019123262250816034,
+      "loss": 1.9868,
+      "step": 760
+    },
+    {
+      "epoch": 0.719122110670091,
+      "grad_norm": 0.03318092492837997,
+      "learning_rate": 0.00019091353021435915,
+      "loss": 1.9943,
+      "step": 770
+    },
+    {
+      "epoch": 0.7284613588606117,
+      "grad_norm": 0.054035272137015325,
+      "learning_rate": 0.00019058901052645844,
+      "loss": 1.9838,
+      "step": 780
+    },
+    {
+      "epoch": 0.7378006070511324,
+      "grad_norm": 0.03184392761983255,
+      "learning_rate": 0.00019025908281777078,
+      "loss": 1.982,
+      "step": 790
+    },
+    {
+      "epoch": 0.747139855241653,
+      "grad_norm": 0.029409948164434735,
+      "learning_rate": 0.00018992376678445908,
+      "loss": 1.9693,
+      "step": 800
+    },
+    {
+      "epoch": 0.7564791034321737,
+      "grad_norm": 0.029656963043919016,
+      "learning_rate": 0.00018958308244436064,
+      "loss": 1.9914,
+      "step": 810
+    },
+    {
+      "epoch": 0.7658183516226944,
+      "grad_norm": 0.030843610865326686,
+      "learning_rate": 0.00018923705013579233,
+      "loss": 1.9749,
+      "step": 820
+    },
+    {
+      "epoch": 0.7751575998132151,
+      "grad_norm": 0.0377584286045999,
+      "learning_rate": 0.00018888569051633613,
+      "loss": 1.9606,
+      "step": 830
+    },
+    {
+      "epoch": 0.7844968480037356,
+      "grad_norm": 0.03147699916274391,
+      "learning_rate": 0.00018852902456160616,
+      "loss": 1.9696,
+      "step": 840
+    },
+    {
+      "epoch": 0.7938360961942563,
+      "grad_norm": 0.029193153251471263,
+      "learning_rate": 0.0001881670735639963,
+      "loss": 1.9687,
+      "step": 850
+    },
+    {
+      "epoch": 0.803175344384777,
+      "grad_norm": 0.030201895228156087,
+      "learning_rate": 0.00018779985913140924,
+      "loss": 1.9678,
+      "step": 860
+    },
+    {
+      "epoch": 0.8125145925752977,
+      "grad_norm": 0.029272319485493213,
+      "learning_rate": 0.00018742740318596632,
+      "loss": 1.9697,
+      "step": 870
+    },
+    {
+      "epoch": 0.8218538407658184,
+      "grad_norm": 0.033740176465285654,
+      "learning_rate": 0.000187049727962699,
+      "loss": 1.9647,
+      "step": 880
+    },
+    {
+      "epoch": 0.831193088956339,
+      "grad_norm": 0.029539399251208593,
+      "learning_rate": 0.0001866668560082213,
+      "loss": 1.9627,
+      "step": 890
+    },
+    {
+      "epoch": 0.8405323371468597,
+      "grad_norm": 0.028666077337951026,
+      "learning_rate": 0.0001862788101793839,
+      "loss": 1.9529,
+      "step": 900
+    },
+    {
+      "epoch": 0.8498715853373804,
+      "grad_norm": 0.030280792492665805,
+      "learning_rate": 0.0001858856136419097,
+      "loss": 1.9536,
+      "step": 910
+    },
+    {
+      "epoch": 0.859210833527901,
+      "grad_norm": 0.03921536105057096,
+      "learning_rate": 0.0001854872898690106,
+      "loss": 1.9474,
+      "step": 920
+    },
+    {
+      "epoch": 0.8685500817184216,
+      "grad_norm": 0.030632523637038354,
+      "learning_rate": 0.0001850838626399865,
+      "loss": 1.9423,
+      "step": 930
+    },
+    {
+      "epoch": 0.8778893299089423,
+      "grad_norm": 0.04615147601979514,
+      "learning_rate": 0.00018467535603880548,
+      "loss": 1.946,
+      "step": 940
+    },
+    {
+      "epoch": 0.887228578099463,
+      "grad_norm": 0.028216236017006333,
+      "learning_rate": 0.00018426179445266616,
+      "loss": 1.9408,
+      "step": 950
+    },
+    {
+      "epoch": 0.8965678262899837,
+      "grad_norm": 0.0282407563402959,
+      "learning_rate": 0.00018384320257054177,
+      "loss": 1.9447,
+      "step": 960
+    },
+    {
+      "epoch": 0.9059070744805043,
+      "grad_norm": 0.029365880854687894,
+      "learning_rate": 0.0001834196053817062,
+      "loss": 1.9389,
+      "step": 970
+    },
+    {
+      "epoch": 0.915246322671025,
+      "grad_norm": 0.02855404439130719,
+      "learning_rate": 0.00018299102817424234,
+      "loss": 1.9425,
+      "step": 980
+    },
+    {
+      "epoch": 0.9245855708615457,
+      "grad_norm": 0.03238310984070135,
+      "learning_rate": 0.00018255749653353225,
+      "loss": 1.9392,
+      "step": 990
+    },
+    {
+      "epoch": 0.9339248190520663,
+      "grad_norm": 0.030115350805430388,
+      "learning_rate": 0.00018211903634072983,
+      "loss": 1.956,
+      "step": 1000
+    },
+    {
+      "epoch": 0.9432640672425869,
+      "grad_norm": 0.039090564032501135,
+      "learning_rate": 0.0001816756737712158,
+      "loss": 1.9358,
+      "step": 1010
+    },
+    {
+      "epoch": 0.9526033154331076,
+      "grad_norm": 0.028567954476327023,
+      "learning_rate": 0.000181227435293035,
+      "loss": 1.9342,
+      "step": 1020
+    },
+    {
+      "epoch": 0.9619425636236283,
+      "grad_norm": 0.028158653241284505,
+      "learning_rate": 0.00018077434766531624,
+      "loss": 1.9287,
+      "step": 1030
+    },
+    {
+      "epoch": 0.971281811814149,
+      "grad_norm": 0.02927795717651538,
+      "learning_rate": 0.00018031643793667504,
+      "loss": 1.9298,
+      "step": 1040
+    },
+    {
+      "epoch": 0.9806210600046696,
+      "grad_norm": 0.06298340584032344,
+      "learning_rate": 0.0001798537334435986,
+      "loss": 1.9303,
+      "step": 1050
+    },
+    {
+      "epoch": 0.9899603081951903,
+      "grad_norm": 0.03627278406983255,
+      "learning_rate": 0.00017938626180881407,
+      "loss": 1.9285,
+      "step": 1060
+    },
+    {
+      "epoch": 0.999299556385711,
+      "grad_norm": 0.03916823421329747,
+      "learning_rate": 0.00017891405093963938,
+      "loss": 1.9239,
+      "step": 1070
+    },
+    {
+      "epoch": 1.0087555451786132,
+      "grad_norm": 0.03383539251115568,
+      "learning_rate": 0.00017843712902631723,
+      "loss": 1.8855,
+      "step": 1080
+    },
+    {
+      "epoch": 1.0180947933691338,
+      "grad_norm": 0.02931021936560147,
+      "learning_rate": 0.00017795552454033224,
+      "loss": 1.8004,
+      "step": 1090
+    },
+    {
+      "epoch": 1.0274340415596543,
+      "grad_norm": 0.029173295095350292,
+      "learning_rate": 0.0001774692662327113,
+      "loss": 1.7912,
+      "step": 1100
+    },
+    {
+      "epoch": 1.0367732897501751,
+      "grad_norm": 0.035210065642974735,
+      "learning_rate": 0.000176978383132307,
+      "loss": 1.7902,
+      "step": 1110
+    },
+    {
+      "epoch": 1.0461125379406957,
+      "grad_norm": 0.029794447659573477,
+      "learning_rate": 0.00017648290454406475,
+      "loss": 1.8072,
+      "step": 1120
+    },
+    {
+      "epoch": 1.0554517861312165,
+      "grad_norm": 0.03194584938279939,
+      "learning_rate": 0.0001759828600472734,
+      "loss": 1.803,
+      "step": 1130
+    },
+    {
+      "epoch": 1.064791034321737,
+      "grad_norm": 0.031028415312581603,
+      "learning_rate": 0.00017547827949379924,
+      "loss": 1.7945,
+      "step": 1140
+    },
+    {
+      "epoch": 1.0741302825122578,
+      "grad_norm": 0.03313245828751237,
+      "learning_rate": 0.00017496919300630403,
+      "loss": 1.8139,
+      "step": 1150
+    },
+    {
+      "epoch": 1.0834695307027784,
+      "grad_norm": 0.03049431724979126,
+      "learning_rate": 0.00017445563097644664,
+      "loss": 1.8031,
+      "step": 1160
+    },
+    {
+      "epoch": 1.0928087788932992,
+      "grad_norm": 0.02880548705343715,
+      "learning_rate": 0.00017393762406306878,
+      "loss": 1.7974,
+      "step": 1170
+    },
+    {
+      "epoch": 1.1021480270838198,
+      "grad_norm": 0.03200427895977668,
+      "learning_rate": 0.00017341520319036469,
+      "loss": 1.7994,
+      "step": 1180
+    },
+    {
+      "epoch": 1.1114872752743403,
+      "grad_norm": 0.031293532815600045,
+      "learning_rate": 0.00017288839954603496,
+      "loss": 1.8073,
+      "step": 1190
+    },
+    {
+      "epoch": 1.1208265234648611,
+      "grad_norm": 0.032367211109345505,
+      "learning_rate": 0.00017235724457942468,
+      "loss": 1.7944,
+      "step": 1200
+    },
+    {
+      "epoch": 1.1301657716553817,
+      "grad_norm": 0.037783793923191374,
+      "learning_rate": 0.0001718217699996462,
+      "loss": 1.7948,
+      "step": 1210
+    },
+    {
+      "epoch": 1.1395050198459025,
+      "grad_norm": 0.02881083117349317,
+      "learning_rate": 0.00017128200777368567,
+      "loss": 1.8029,
+      "step": 1220
+    },
+    {
+      "epoch": 1.148844268036423,
+      "grad_norm": 0.032997408408175985,
+      "learning_rate": 0.00017073799012449524,
+      "loss": 1.7914,
+      "step": 1230
+    },
+    {
+      "epoch": 1.1581835162269438,
+      "grad_norm": 0.030263615801190885,
+      "learning_rate": 0.00017018974952906884,
+      "loss": 1.792,
+      "step": 1240
+    },
+    {
+      "epoch": 1.1675227644174644,
+      "grad_norm": 0.03286310429098484,
+      "learning_rate": 0.00016963731871650378,
+      "loss": 1.8149,
+      "step": 1250
+    },
+    {
+      "epoch": 1.1768620126079852,
+      "grad_norm": 0.031111621587597126,
+      "learning_rate": 0.00016908073066604663,
+      "loss": 1.8092,
+      "step": 1260
+    },
+    {
+      "epoch": 1.1862012607985057,
+      "grad_norm": 0.03372950172075241,
+      "learning_rate": 0.0001685200186051246,
+      "loss": 1.818,
+      "step": 1270
+    },
+    {
+      "epoch": 1.1955405089890263,
+      "grad_norm": 0.040782040436567434,
+      "learning_rate": 0.00016795521600736164,
+      "loss": 1.7999,
+      "step": 1280
+    },
+    {
+      "epoch": 1.204879757179547,
+      "grad_norm": 0.03451605168178924,
+      "learning_rate": 0.00016738635659058044,
+      "loss": 1.7945,
+      "step": 1290
+    },
+    {
+      "epoch": 1.2142190053700677,
+      "grad_norm": 0.03235681588882673,
+      "learning_rate": 0.00016681347431478933,
+      "loss": 1.8087,
+      "step": 1300
+    },
+    {
+      "epoch": 1.2235582535605884,
+      "grad_norm": 0.030750745605971932,
+      "learning_rate": 0.00016623660338015487,
+      "loss": 1.7995,
+      "step": 1310
+    },
+    {
+      "epoch": 1.232897501751109,
+      "grad_norm": 0.029444668665577274,
+      "learning_rate": 0.00016565577822496042,
+      "loss": 1.8025,
+      "step": 1320
+    },
+    {
+      "epoch": 1.2422367499416298,
+      "grad_norm": 0.038528856709584745,
+      "learning_rate": 0.00016507103352354996,
+      "loss": 1.7954,
+      "step": 1330
+    },
+    {
+      "epoch": 1.2515759981321504,
+      "grad_norm": 0.034217088004383035,
+      "learning_rate": 0.00016448240418425814,
+      "loss": 1.7962,
+      "step": 1340
+    },
+    {
+      "epoch": 1.260915246322671,
+      "grad_norm": 0.030205405393195585,
+      "learning_rate": 0.00016388992534732645,
+      "loss": 1.7973,
+      "step": 1350
+    },
+    {
+      "epoch": 1.2702544945131917,
+      "grad_norm": 0.029082218516562994,
+      "learning_rate": 0.00016329363238280528,
+      "loss": 1.796,
+      "step": 1360
+    },
+    {
+      "epoch": 1.2795937427037123,
+      "grad_norm": 0.029003887688766505,
+      "learning_rate": 0.00016269356088844238,
+      "loss": 1.7946,
+      "step": 1370
+    },
+    {
+      "epoch": 1.288932990894233,
+      "grad_norm": 0.03341157363649238,
+      "learning_rate": 0.00016208974668755779,
+      "loss": 1.7972,
+      "step": 1380
+    },
+    {
+      "epoch": 1.2982722390847536,
+      "grad_norm": 0.030614480844663026,
+      "learning_rate": 0.00016148222582690517,
+      "loss": 1.7973,
+      "step": 1390
+    },
+    {
+      "epoch": 1.3076114872752744,
+      "grad_norm": 0.029741346740467405,
+      "learning_rate": 0.00016087103457452,
+      "loss": 1.8076,
+      "step": 1400
+    },
+    {
+      "epoch": 1.316950735465795,
+      "grad_norm": 0.029569313554185597,
+      "learning_rate": 0.00016025620941755424,
+      "loss": 1.8043,
+      "step": 1410
+    },
+    {
+      "epoch": 1.3262899836563156,
+      "grad_norm": 0.02947637404374054,
+      "learning_rate": 0.0001596377870600983,
+      "loss": 1.797,
+      "step": 1420
+    },
+    {
+      "epoch": 1.3356292318468364,
+      "grad_norm": 0.031005062093959545,
+      "learning_rate": 0.00015901580442098968,
+      "loss": 1.8086,
+      "step": 1430
+    },
+    {
+      "epoch": 1.344968480037357,
+      "grad_norm": 0.029493792984873927,
+      "learning_rate": 0.00015839029863160922,
+      "loss": 1.8026,
+      "step": 1440
+    },
+    {
+      "epoch": 1.3543077282278777,
+      "grad_norm": 0.0288068155951218,
+      "learning_rate": 0.0001577613070336641,
+      "loss": 1.7951,
+      "step": 1450
+    },
+    {
+      "epoch": 1.3636469764183983,
+      "grad_norm": 0.03380404824627639,
+      "learning_rate": 0.00015712886717695885,
+      "loss": 1.7938,
+      "step": 1460
+    },
+    {
+      "epoch": 1.372986224608919,
+      "grad_norm": 0.039744706189693335,
+      "learning_rate": 0.0001564930168171536,
+      "loss": 1.8016,
+      "step": 1470
+    },
+    {
+      "epoch": 1.3823254727994396,
+      "grad_norm": 0.030565530594285437,
+      "learning_rate": 0.00015585379391351012,
+      "loss": 1.7984,
+      "step": 1480
+    },
+    {
+      "epoch": 1.3916647209899602,
+      "grad_norm": 0.04009392805554255,
+      "learning_rate": 0.00015521123662662567,
+      "loss": 1.7999,
+      "step": 1490
+    },
+    {
+      "epoch": 1.401003969180481,
+      "grad_norm": 0.03516196009586836,
+      "learning_rate": 0.000154565383316155,
+      "loss": 1.7979,
+      "step": 1500
+    },
+    {
+      "epoch": 1.4103432173710018,
+      "grad_norm": 0.03534161399054556,
+      "learning_rate": 0.0001539162725385202,
+      "loss": 1.8057,
+      "step": 1510
+    },
+    {
+      "epoch": 1.4196824655615223,
+      "grad_norm": 0.028488879438601067,
+      "learning_rate": 0.000153263943044609,
+      "loss": 1.792,
+      "step": 1520
+    },
+    {
+      "epoch": 1.429021713752043,
+      "grad_norm": 0.03125154490954804,
+      "learning_rate": 0.00015260843377746147,
+      "loss": 1.8008,
+      "step": 1530
+    },
+    {
+      "epoch": 1.4383609619425637,
+      "grad_norm": 0.030194357488801882,
+      "learning_rate": 0.00015194978386994507,
+      "loss": 1.7948,
+      "step": 1540
+    },
+    {
+      "epoch": 1.4477002101330843,
+      "grad_norm": 0.03049246845786265,
+      "learning_rate": 0.00015128803264241852,
+      "loss": 1.7967,
+      "step": 1550
+    },
+    {
+      "epoch": 1.4570394583236048,
+      "grad_norm": 0.030497211097258083,
+      "learning_rate": 0.0001506232196003844,
+      "loss": 1.7894,
+      "step": 1560
+    },
+    {
+      "epoch": 1.4663787065141256,
+      "grad_norm": 0.028748806119737205,
+      "learning_rate": 0.00014995538443213094,
+      "loss": 1.806,
+      "step": 1570
+    },
+    {
+      "epoch": 1.4757179547046464,
+      "grad_norm": 0.036423750322912396,
+      "learning_rate": 0.00014928456700636237,
+      "loss": 1.7995,
+      "step": 1580
+    },
+    {
+      "epoch": 1.485057202895167,
+      "grad_norm": 0.039101516109204065,
+      "learning_rate": 0.00014861080736981906,
+      "loss": 1.8028,
+      "step": 1590
+    },
+    {
+      "epoch": 1.4943964510856875,
+      "grad_norm": 0.031368399541673815,
+      "learning_rate": 0.00014793414574488663,
+      "loss": 1.8005,
+      "step": 1600
+    },
+    {
+      "epoch": 1.5037356992762083,
+      "grad_norm": 0.029788484702512056,
+      "learning_rate": 0.00014725462252719495,
+      "loss": 1.7963,
+      "step": 1610
+    },
+    {
+      "epoch": 1.5130749474667289,
+      "grad_norm": 0.029719041811636312,
+      "learning_rate": 0.00014657227828320635,
+      "loss": 1.7957,
+      "step": 1620
+    },
+    {
+      "epoch": 1.5224141956572494,
+      "grad_norm": 0.02820041575417432,
+      "learning_rate": 0.00014588715374779407,
+      "loss": 1.7986,
+      "step": 1630
+    },
+    {
+      "epoch": 1.5317534438477702,
+      "grad_norm": 1.8322544897261024,
+      "learning_rate": 0.0001451992898218102,
+      "loss": 2.016,
+      "step": 1640
+    },
+    {
+      "epoch": 1.541092692038291,
+      "grad_norm": 0.8676995793107466,
+      "learning_rate": 0.0001445087275696443,
+      "loss": 2.1095,
+      "step": 1650
+    },
+    {
+      "epoch": 1.5504319402288116,
+      "grad_norm": 0.08688193434631736,
+      "learning_rate": 0.00014381550821677155,
+      "loss": 2.0497,
+      "step": 1660
+    },
+    {
+      "epoch": 1.5597711884193322,
+      "grad_norm": 0.18321381224589608,
+      "learning_rate": 0.0001431196731472921,
+      "loss": 1.9023,
+      "step": 1670
+    },
+    {
+      "epoch": 1.569110436609853,
+      "grad_norm": 0.058214343698110564,
+      "learning_rate": 0.00014242126390145998,
+      "loss": 1.8566,
+      "step": 1680
+    },
+    {
+      "epoch": 1.5784496848003737,
+      "grad_norm": 0.03965578147557666,
+      "learning_rate": 0.0001417203221732036,
+      "loss": 1.8206,
+      "step": 1690
+    },
+    {
+      "epoch": 1.587788932990894,
+      "grad_norm": 0.03131802880017099,
+      "learning_rate": 0.00014101688980763658,
+      "loss": 1.8272,
+      "step": 1700
+    },
+    {
+      "epoch": 1.5971281811814149,
+      "grad_norm": 0.02788722706683908,
+      "learning_rate": 0.00014031100879855968,
+      "loss": 1.8145,
+      "step": 1710
+    },
+    {
+      "epoch": 1.6064674293719357,
+      "grad_norm": 0.02781057130092059,
+      "learning_rate": 0.00013960272128595372,
+      "loss": 1.8122,
+      "step": 1720
+    },
+    {
+      "epoch": 1.6158066775624562,
+      "grad_norm": 0.032067383734154756,
+      "learning_rate": 0.00013889206955346403,
+      "loss": 1.8064,
+      "step": 1730
+    },
+    {
+      "epoch": 1.6251459257529768,
+      "grad_norm": 0.03448807133884858,
+      "learning_rate": 0.00013817909602587613,
+      "loss": 1.8128,
+      "step": 1740
+    },
+    {
+      "epoch": 1.6344851739434976,
+      "grad_norm": 0.04127395288271696,
+      "learning_rate": 0.00013746384326658305,
+      "loss": 1.8041,
+      "step": 1750
+    },
+    {
+      "epoch": 1.6438244221340184,
+      "grad_norm": 0.030243192787820814,
+      "learning_rate": 0.00013674635397504427,
+      "loss": 1.803,
+      "step": 1760
+    },
+    {
+      "epoch": 1.653163670324539,
+      "grad_norm": 0.032183525625428915,
+      "learning_rate": 0.00013602667098423687,
+      "loss": 1.8092,
+      "step": 1770
+    },
+    {
+      "epoch": 1.6625029185150595,
+      "grad_norm": 0.030391044676815377,
+      "learning_rate": 0.00013530483725809818,
+      "loss": 1.8039,
+      "step": 1780
+    },
+    {
+      "epoch": 1.6718421667055803,
+      "grad_norm": 0.03204756818238517,
+      "learning_rate": 0.000134580895888961,
+      "loss": 1.8017,
+      "step": 1790
+    },
+    {
+      "epoch": 1.6811814148961008,
+      "grad_norm": 0.031117112662107062,
+      "learning_rate": 0.00013385489009498124,
+      "loss": 1.8017,
+      "step": 1800
+    },
+    {
+      "epoch": 1.6905206630866214,
+      "grad_norm": 0.028389293445401805,
+      "learning_rate": 0.00013312686321755761,
+      "loss": 1.811,
+      "step": 1810
+    },
+    {
+      "epoch": 1.6998599112771422,
+      "grad_norm": 4.908185097372493,
+      "learning_rate": 0.0001323968587187443,
+      "loss": 2.194,
+      "step": 1820
+    },
+    {
+      "epoch": 1.709199159467663,
+      "grad_norm": 0.2849452041194025,
+      "learning_rate": 0.00013166492017865637,
+      "loss": 2.0785,
+      "step": 1830
+    },
+    {
+      "epoch": 1.7185384076581836,
+      "grad_norm": 0.10989252058989733,
+      "learning_rate": 0.0001309310912928682,
+      "loss": 1.986,
+      "step": 1840
+    },
+    {
+      "epoch": 1.7278776558487041,
+      "grad_norm": 0.057475612656740484,
+      "learning_rate": 0.00013019541586980463,
+      "loss": 1.8614,
+      "step": 1850
+    },
+    {
+      "epoch": 1.737216904039225,
+      "grad_norm": 0.034908181734207726,
+      "learning_rate": 0.000129457937828126,
+      "loss": 1.8326,
+      "step": 1860
+    },
+    {
+      "epoch": 1.7465561522297455,
+      "grad_norm": 0.02892836681897248,
+      "learning_rate": 0.00012871870119410614,
+      "loss": 1.8243,
+      "step": 1870
+    },
+    {
+      "epoch": 1.755895400420266,
+      "grad_norm": 0.03311636981729384,
+      "learning_rate": 0.00012797775009900397,
+      "loss": 1.8183,
+      "step": 1880
+    },
+    {
+      "epoch": 1.7652346486107868,
+      "grad_norm": 0.03151917952358458,
+      "learning_rate": 0.00012723512877642904,
+      "loss": 1.8034,
+      "step": 1890
+    },
+    {
+      "epoch": 1.7745738968013076,
+      "grad_norm": 0.028109921832296925,
+      "learning_rate": 0.000126490881559701,
+      "loss": 1.8129,
+      "step": 1900
+    },
+    {
+      "epoch": 1.7839131449918282,
+      "grad_norm": 0.030350462454962698,
+      "learning_rate": 0.00012574505287920259,
+      "loss": 1.8003,
+      "step": 1910
+    },
+    {
+      "epoch": 1.7932523931823487,
+      "grad_norm": 0.03131380630103849,
+      "learning_rate": 0.00012499768725972754,
+      "loss": 1.814,
+      "step": 1920
+    },
+    {
+      "epoch": 1.8025916413728695,
+      "grad_norm": 0.029450198273050322,
+      "learning_rate": 0.00012424882931782243,
+      "loss": 1.7998,
+      "step": 1930
+    },
+    {
+      "epoch": 1.81193088956339,
+      "grad_norm": 0.0310524261453681,
+      "learning_rate": 0.0001234985237591231,
+      "loss": 1.8078,
+      "step": 1940
+    },
+    {
+      "epoch": 1.8212701377539107,
+      "grad_norm": 0.029362038478982822,
+      "learning_rate": 0.00012274681537568585,
+      "loss": 1.8014,
+      "step": 1950
+    },
+    {
+      "epoch": 1.8306093859444315,
+      "grad_norm": 0.027166816460226118,
+      "learning_rate": 0.00012199374904331337,
+      "loss": 1.8021,
+      "step": 1960
+    },
+    {
+      "epoch": 1.8399486341349522,
+      "grad_norm": 0.03680509283276228,
+      "learning_rate": 0.00012123936971887578,
+      "loss": 1.7973,
+      "step": 1970
+    },
+    {
+      "epoch": 1.8492878823254728,
+      "grad_norm": 0.03135174840346185,
+      "learning_rate": 0.0001204837224376267,
+      "loss": 1.7874,
+      "step": 1980
+    },
+    {
+      "epoch": 1.8586271305159934,
+      "grad_norm": 0.02644533256389969,
+      "learning_rate": 0.0001197268523105148,
+      "loss": 1.798,
+      "step": 1990
+    },
+    {
+      "epoch": 1.8679663787065142,
+      "grad_norm": 0.02999453651649614,
+      "learning_rate": 0.00011896880452149077,
+      "loss": 1.7957,
+      "step": 2000
+    },
+    {
+      "epoch": 1.877305626897035,
+      "grad_norm": 0.026905209700322272,
+      "learning_rate": 0.00011820962432480985,
+      "loss": 1.793,
+      "step": 2010
+    },
+    {
+      "epoch": 1.8866448750875553,
+      "grad_norm": 0.027263640323285022,
+      "learning_rate": 0.00011744935704233005,
+      "loss": 1.7974,
+      "step": 2020
+    },
+    {
+      "epoch": 1.895984123278076,
+      "grad_norm": 0.030479226063932337,
+      "learning_rate": 0.00011668804806080693,
+      "loss": 1.7898,
+      "step": 2030
+    },
+    {
+      "epoch": 1.9053233714685969,
+      "grad_norm": 0.030129902025534238,
+      "learning_rate": 0.00011592574282918369,
+      "loss": 1.7856,
+      "step": 2040
+    },
+    {
+      "epoch": 1.9146626196591174,
+      "grad_norm": 0.027884976674153635,
+      "learning_rate": 0.00011516248685587814,
+      "loss": 1.7858,
+      "step": 2050
+    },
+    {
+      "epoch": 1.924001867849638,
+      "grad_norm": 0.02925266011156687,
+      "learning_rate": 0.00011439832570606586,
+      "loss": 1.7876,
+      "step": 2060
+    },
+    {
+      "epoch": 1.9333411160401588,
+      "grad_norm": 0.028472914828616754,
+      "learning_rate": 0.00011363330499895997,
+      "loss": 1.7834,
+      "step": 2070
+    },
+    {
+      "epoch": 1.9426803642306796,
+      "grad_norm": 0.025877740032137875,
+      "learning_rate": 0.00011286747040508789,
+      "loss": 1.7955,
+      "step": 2080
+    },
+    {
+      "epoch": 1.9520196124212001,
+      "grad_norm": 0.02605295620697312,
+      "learning_rate": 0.0001121008676435648,
+      "loss": 1.7877,
+      "step": 2090
+    },
+    {
+      "epoch": 1.9613588606117207,
+      "grad_norm": 0.026887649929567867,
+      "learning_rate": 0.00011133354247936423,
+      "loss": 1.773,
+      "step": 2100
+    },
+    {
+      "epoch": 1.9706981088022415,
+      "grad_norm": 0.027982045915154026,
+      "learning_rate": 0.00011056554072058596,
+      "loss": 1.7762,
+      "step": 2110
+    },
+    {
+      "epoch": 1.980037356992762,
+      "grad_norm": 0.028077027059053006,
+      "learning_rate": 0.0001097969082157215,
+      "loss": 1.7963,
+      "step": 2120
+    },
+    {
+      "epoch": 1.9893766051832826,
+      "grad_norm": 0.027877078975954036,
+      "learning_rate": 0.00010902769085091686,
+      "loss": 1.7787,
+      "step": 2130
+    },
+    {
+      "epoch": 1.9987158533738034,
+      "grad_norm": 0.026120077388738373,
+      "learning_rate": 0.00010825793454723325,
+      "loss": 1.7842,
+      "step": 2140
+    },
+    {
+      "epoch": 2.0081718421667056,
+      "grad_norm": 0.044810283079268924,
+      "learning_rate": 0.00010748768525790569,
+      "loss": 1.6591,
+      "step": 2150
+    },
+    {
+      "epoch": 2.0175110903572264,
+      "grad_norm": 0.03431848280739808,
+      "learning_rate": 0.00010671698896559968,
+      "loss": 1.5599,
+      "step": 2160
+    },
+    {
+      "epoch": 2.0268503385477468,
+      "grad_norm": 0.04234332973849956,
+      "learning_rate": 0.00010594589167966606,
+      "loss": 1.5494,
+      "step": 2170
+    },
+    {
+      "epoch": 2.0361895867382676,
+      "grad_norm": 0.03260321438171042,
+      "learning_rate": 0.00010517443943339438,
+      "loss": 1.5473,
+      "step": 2180
+    },
+    {
+      "epoch": 2.0455288349287883,
+      "grad_norm": 0.034116901745609114,
+      "learning_rate": 0.00010440267828126478,
+      "loss": 1.5464,
+      "step": 2190
+    },
+    {
+      "epoch": 2.0548680831193087,
+      "grad_norm": 0.030992757239375807,
+      "learning_rate": 0.00010363065429619858,
+      "loss": 1.5514,
+      "step": 2200
+    },
+    {
+      "epoch": 2.0642073313098295,
+      "grad_norm": 0.03365516197786113,
+      "learning_rate": 0.0001028584135668077,
+      "loss": 1.5493,
+      "step": 2210
+    },
+    {
+      "epoch": 2.0735465795003503,
+      "grad_norm": 0.033293307482261586,
+      "learning_rate": 0.00010208600219464355,
+      "loss": 1.5426,
+      "step": 2220
+    },
+    {
+      "epoch": 2.082885827690871,
+      "grad_norm": 0.03653097737467338,
+      "learning_rate": 0.00010131346629144451,
+      "loss": 1.5471,
+      "step": 2230
+    },
+    {
+      "epoch": 2.0922250758813914,
+      "grad_norm": 0.03390291511697895,
+      "learning_rate": 0.0001005408519763833,
+      "loss": 1.5568,
+      "step": 2240
+    },
+    {
+      "epoch": 2.101564324071912,
+      "grad_norm": 0.03192694661852283,
+      "learning_rate": 9.976820537331374e-05,
+      "loss": 1.5452,
+      "step": 2250
+    },
+    {
+      "epoch": 2.110903572262433,
+      "grad_norm": 0.03561740193515691,
+      "learning_rate": 9.899557260801707e-05,
+      "loss": 1.546,
+      "step": 2260
+    },
+    {
+      "epoch": 2.1202428204529538,
+      "grad_norm": 0.029803732953068658,
+      "learning_rate": 9.822299980544862e-05,
+      "loss": 1.5533,
+      "step": 2270
+    },
+    {
+      "epoch": 2.129582068643474,
+      "grad_norm": 0.031232417271289125,
+      "learning_rate": 9.745053308698392e-05,
+      "loss": 1.5469,
+      "step": 2280
+    },
+    {
+      "epoch": 2.138921316833995,
+      "grad_norm": 0.032434793780181034,
+      "learning_rate": 9.667821856766548e-05,
+      "loss": 1.5514,
+      "step": 2290
+    },
+    {
+      "epoch": 2.1482605650245157,
+      "grad_norm": 0.03579370906405582,
+      "learning_rate": 9.590610235344972e-05,
+      "loss": 1.5577,
+      "step": 2300
+    },
+    {
+      "epoch": 2.157599813215036,
+      "grad_norm": 0.029662202478648328,
+      "learning_rate": 9.51342305384546e-05,
+      "loss": 1.5543,
+      "step": 2310
+    },
+    {
+      "epoch": 2.166939061405557,
+      "grad_norm": 0.03178715913934592,
+      "learning_rate": 9.436264920220781e-05,
+      "loss": 1.5579,
+      "step": 2320
+    },
+    {
+      "epoch": 2.1762783095960776,
+      "grad_norm": 0.03384008887051677,
+      "learning_rate": 9.359140440689601e-05,
+      "loss": 1.5595,
+      "step": 2330
+    },
+    {
+      "epoch": 2.1856175577865984,
+      "grad_norm": 0.03316450664408166,
+      "learning_rate": 9.282054219461475e-05,
+      "loss": 1.5556,
+      "step": 2340
+    },
+    {
+      "epoch": 2.1949568059771187,
+      "grad_norm": 0.032176305552558876,
+      "learning_rate": 9.205010858462007e-05,
+      "loss": 1.5638,
+      "step": 2350
+    },
+    {
+      "epoch": 2.2042960541676395,
+      "grad_norm": 0.031134335256756362,
+      "learning_rate": 9.128014957058109e-05,
+      "loss": 1.5629,
+      "step": 2360
+    },
+    {
+      "epoch": 2.2136353023581603,
+      "grad_norm": 0.032205851810441756,
+      "learning_rate": 9.051071111783436e-05,
+      "loss": 1.5613,
+      "step": 2370
+    },
+    {
+      "epoch": 2.2229745505486807,
+      "grad_norm": 0.029048245476020442,
+      "learning_rate": 8.974183916063968e-05,
+      "loss": 1.5594,
+      "step": 2380
+    },
+    {
+      "epoch": 2.2323137987392014,
+      "grad_norm": 0.03183166054573621,
+      "learning_rate": 8.897357959943795e-05,
+      "loss": 1.5606,
+      "step": 2390
+    },
+    {
+      "epoch": 2.2416530469297222,
+      "grad_norm": 0.032309922117136916,
+      "learning_rate": 8.820597829811109e-05,
+      "loss": 1.5524,
+      "step": 2400
+    },
+    {
+      "epoch": 2.250992295120243,
+      "grad_norm": 0.03598922231958808,
+      "learning_rate": 8.743908108124388e-05,
+      "loss": 1.5604,
+      "step": 2410
+    },
+    {
+      "epoch": 2.2603315433107634,
+      "grad_norm": 0.03046424735786346,
+      "learning_rate": 8.667293373138835e-05,
+      "loss": 1.5598,
+      "step": 2420
+    },
+    {
+      "epoch": 2.269670791501284,
+      "grad_norm": 0.030995453538377543,
+      "learning_rate": 8.59075819863307e-05,
+      "loss": 1.5652,
+      "step": 2430
+    },
+    {
+      "epoch": 2.279010039691805,
+      "grad_norm": 0.029309020623010097,
+      "learning_rate": 8.514307153636077e-05,
+      "loss": 1.5651,
+      "step": 2440
+    },
+    {
+      "epoch": 2.2883492878823253,
+      "grad_norm": 0.03158721106736763,
+      "learning_rate": 8.437944802154434e-05,
+      "loss": 1.5581,
+      "step": 2450
+    },
+    {
+      "epoch": 2.297688536072846,
+      "grad_norm": 0.03168229084049938,
+      "learning_rate": 8.361675702899871e-05,
+      "loss": 1.5671,
+      "step": 2460
+    },
+    {
+      "epoch": 2.307027784263367,
+      "grad_norm": 0.031335222148495136,
+      "learning_rate": 8.2855044090171e-05,
+      "loss": 1.5675,
+      "step": 2470
+    },
+    {
+      "epoch": 2.3163670324538876,
+      "grad_norm": 0.031073847325941303,
+      "learning_rate": 8.209435467811998e-05,
+      "loss": 1.5624,
+      "step": 2480
+    },
+    {
+      "epoch": 2.325706280644408,
+      "grad_norm": 0.030099100631045844,
+      "learning_rate": 8.133473420480161e-05,
+      "loss": 1.5606,
+      "step": 2490
+    },
+    {
+      "epoch": 2.3350455288349288,
+      "grad_norm": 0.034147632950361176,
+      "learning_rate": 8.057622801835788e-05,
+      "loss": 1.5703,
+      "step": 2500
+    },
+    {
+      "epoch": 2.3443847770254496,
+      "grad_norm": 0.03051580784550685,
+      "learning_rate": 7.981888140040955e-05,
+      "loss": 1.5731,
+      "step": 2510
+    },
+    {
+      "epoch": 2.3537240252159704,
+      "grad_norm": 0.03068917065832597,
+      "learning_rate": 7.9062739563353e-05,
+      "loss": 1.5723,
+      "step": 2520
+    },
+    {
+      "epoch": 2.3630632734064907,
+      "grad_norm": 0.02899547641705554,
+      "learning_rate": 7.830784764766118e-05,
+      "loss": 1.5691,
+      "step": 2530
+    },
+    {
+      "epoch": 2.3724025215970115,
+      "grad_norm": 0.030965383166701443,
+      "learning_rate": 7.755425071918858e-05,
+      "loss": 1.5627,
+      "step": 2540
+    },
+    {
+      "epoch": 2.3817417697875323,
+      "grad_norm": 0.03252440018336625,
+      "learning_rate": 7.680199376648108e-05,
+      "loss": 1.5536,
+      "step": 2550
+    },
+    {
+      "epoch": 2.3910810179780526,
+      "grad_norm": 0.031720485449340044,
+      "learning_rate": 7.605112169809008e-05,
+      "loss": 1.5617,
+      "step": 2560
+    },
+    {
+      "epoch": 2.4004202661685734,
+      "grad_norm": 0.031796658969132544,
+      "learning_rate": 7.530167933989161e-05,
+      "loss": 1.5595,
+      "step": 2570
+    },
+    {
+      "epoch": 2.409759514359094,
+      "grad_norm": 0.03218288097429844,
+      "learning_rate": 7.45537114324102e-05,
+      "loss": 1.5628,
+      "step": 2580
+    },
+    {
+      "epoch": 2.4190987625496145,
+      "grad_norm": 0.0305713183075559,
+      "learning_rate": 7.380726262814814e-05,
+      "loss": 1.5717,
+      "step": 2590
+    },
+    {
+      "epoch": 2.4284380107401353,
+      "grad_norm": 0.06879342166341705,
+      "learning_rate": 7.30623774889195e-05,
+      "loss": 1.5726,
+      "step": 2600
+    },
+    {
+      "epoch": 2.437777258930656,
+      "grad_norm": 0.04101428600237338,
+      "learning_rate": 7.231910048319011e-05,
+      "loss": 1.5679,
+      "step": 2610
+    },
+    {
+      "epoch": 2.447116507121177,
+      "grad_norm": 0.031060002638443395,
+      "learning_rate": 7.157747598342274e-05,
+      "loss": 1.562,
+      "step": 2620
+    },
+    {
+      "epoch": 2.4564557553116972,
+      "grad_norm": 0.032302829437386466,
+      "learning_rate": 7.083754826342816e-05,
+      "loss": 1.5767,
+      "step": 2630
+    },
+    {
+      "epoch": 2.465795003502218,
+      "grad_norm": 0.03111462744196413,
+      "learning_rate": 7.009936149572205e-05,
+      "loss": 1.5672,
+      "step": 2640
+    },
+    {
+      "epoch": 2.475134251692739,
+      "grad_norm": 0.031134083521743777,
+      "learning_rate": 6.936295974888807e-05,
+      "loss": 1.5665,
+      "step": 2650
+    },
+    {
+      "epoch": 2.4844734998832596,
+      "grad_norm": 0.030961556373721985,
+      "learning_rate": 6.862838698494693e-05,
+      "loss": 1.5608,
+      "step": 2660
+    },
+    {
+      "epoch": 2.49381274807378,
+      "grad_norm": 0.03168121700432082,
+      "learning_rate": 6.789568705673183e-05,
+      "loss": 1.566,
+      "step": 2670
+    },
+    {
+      "epoch": 2.5031519962643007,
+      "grad_norm": 0.030850372541726772,
+      "learning_rate": 6.716490370527081e-05,
+      "loss": 1.5651,
+      "step": 2680
+    },
+    {
+      "epoch": 2.5124912444548215,
+      "grad_norm": 0.03076635908430861,
+      "learning_rate": 6.643608055717519e-05,
+      "loss": 1.5596,
+      "step": 2690
+    },
+    {
+      "epoch": 2.521830492645342,
+      "grad_norm": 0.031897253741779714,
+      "learning_rate": 6.570926112203528e-05,
+      "loss": 1.5716,
+      "step": 2700
+    },
+    {
+      "epoch": 2.5311697408358627,
+      "grad_norm": 0.03085546721246857,
+      "learning_rate": 6.498448878982291e-05,
+      "loss": 1.5647,
+      "step": 2710
+    },
+    {
+      "epoch": 2.5405089890263834,
+      "grad_norm": 0.03127518794548787,
+      "learning_rate": 6.426180682830107e-05,
+      "loss": 1.5573,
+      "step": 2720
+    },
+    {
+      "epoch": 2.549848237216904,
+      "grad_norm": 0.03196649247686066,
+      "learning_rate": 6.354125838044098e-05,
+      "loss": 1.5597,
+      "step": 2730
+    },
+    {
+      "epoch": 2.5591874854074246,
+      "grad_norm": 0.030359755432035333,
+      "learning_rate": 6.282288646184638e-05,
+      "loss": 1.5625,
+      "step": 2740
+    },
+    {
+      "epoch": 2.5685267335979454,
+      "grad_norm": 0.03030640438940187,
+      "learning_rate": 6.210673395818571e-05,
+      "loss": 1.5717,
+      "step": 2750
+    },
+    {
+      "epoch": 2.577865981788466,
+      "grad_norm": 0.032197470232298186,
+      "learning_rate": 6.139284362263185e-05,
+      "loss": 1.5663,
+      "step": 2760
+    },
+    {
+      "epoch": 2.587205229978987,
+      "grad_norm": 0.030983733397891462,
+      "learning_rate": 6.0681258073309756e-05,
+      "loss": 1.5657,
+      "step": 2770
+    },
+    {
+      "epoch": 2.5965444781695073,
+      "grad_norm": 0.030427577702286164,
+      "learning_rate": 5.9972019790752385e-05,
+      "loss": 1.5708,
+      "step": 2780
+    },
+    {
+      "epoch": 2.605883726360028,
+      "grad_norm": 0.032761226318855745,
+      "learning_rate": 5.9265171115364495e-05,
+      "loss": 1.5641,
+      "step": 2790
+    },
+    {
+      "epoch": 2.615222974550549,
+      "grad_norm": 0.0317533648622182,
+      "learning_rate": 5.856075424489511e-05,
+      "loss": 1.5613,
+      "step": 2800
+    },
+    {
+      "epoch": 2.624562222741069,
+      "grad_norm": 0.03251887796623927,
+      "learning_rate": 5.785881123191834e-05,
+      "loss": 1.5644,
+      "step": 2810
+    },
+    {
+      "epoch": 2.63390147093159,
+      "grad_norm": 0.033109556138937665,
+      "learning_rate": 5.7159383981322866e-05,
+      "loss": 1.5613,
+      "step": 2820
+    },
+    {
+      "epoch": 2.643240719122111,
+      "grad_norm": 0.03190494841434546,
+      "learning_rate": 5.646251424781044e-05,
+      "loss": 1.5597,
+      "step": 2830
+    },
+    {
+      "epoch": 2.652579967312631,
+      "grad_norm": 0.03131425467271855,
+      "learning_rate": 5.576824363340293e-05,
+      "loss": 1.5644,
+      "step": 2840
+    },
+    {
+      "epoch": 2.661919215503152,
+      "grad_norm": 0.032580122114687284,
+      "learning_rate": 5.507661358495904e-05,
+      "loss": 1.5651,
+      "step": 2850
+    },
+    {
+      "epoch": 2.6712584636936727,
+      "grad_norm": 0.031118294785578867,
+      "learning_rate": 5.4387665391699814e-05,
+      "loss": 1.5595,
+      "step": 2860
+    },
+    {
+      "epoch": 2.680597711884193,
+      "grad_norm": 0.030454480461742643,
+      "learning_rate": 5.370144018274371e-05,
+      "loss": 1.5607,
+      "step": 2870
+    },
+    {
+      "epoch": 2.689936960074714,
+      "grad_norm": 0.030828671799046907,
+      "learning_rate": 5.301797892465148e-05,
+      "loss": 1.5587,
+      "step": 2880
+    },
+    {
+      "epoch": 2.6992762082652346,
+      "grad_norm": 0.030541588338347705,
+      "learning_rate": 5.2337322418980204e-05,
+      "loss": 1.5698,
+      "step": 2890
+    },
+    {
+      "epoch": 2.7086154564557554,
+      "grad_norm": 0.030634684731728472,
+      "learning_rate": 5.16595112998477e-05,
+      "loss": 1.5628,
+      "step": 2900
+    },
+    {
+      "epoch": 2.717954704646276,
+      "grad_norm": 0.03045168911848818,
+      "learning_rate": 5.098458603150691e-05,
+      "loss": 1.5544,
+      "step": 2910
+    },
+    {
+      "epoch": 2.7272939528367965,
+      "grad_norm": 0.029966073421542574,
+      "learning_rate": 5.0312586905929816e-05,
+      "loss": 1.557,
+      "step": 2920
+    },
+    {
+      "epoch": 2.7366332010273173,
+      "grad_norm": 0.031203173596988765,
+      "learning_rate": 4.964355404040232e-05,
+      "loss": 1.5571,
+      "step": 2930
+    },
+    {
+      "epoch": 2.745972449217838,
+      "grad_norm": 0.02933060802902336,
+      "learning_rate": 4.897752737512944e-05,
+      "loss": 1.5518,
+      "step": 2940
+    },
+    {
+      "epoch": 2.7553116974083585,
+      "grad_norm": 0.03039237943996916,
+      "learning_rate": 4.8314546670850594e-05,
+      "loss": 1.5682,
+      "step": 2950
+    },
+    {
+      "epoch": 2.7646509455988793,
+      "grad_norm": 0.03077504425768828,
+      "learning_rate": 4.765465150646633e-05,
+      "loss": 1.5618,
+      "step": 2960
+    },
+    {
+      "epoch": 2.7739901937894,
+      "grad_norm": 0.02987361489256037,
+      "learning_rate": 4.699788127667517e-05,
+      "loss": 1.5657,
+      "step": 2970
+    },
+    {
+      "epoch": 2.7833294419799204,
+      "grad_norm": 0.030823634404763923,
+      "learning_rate": 4.634427518962209e-05,
+      "loss": 1.5611,
+      "step": 2980
+    },
+    {
+      "epoch": 2.792668690170441,
+      "grad_norm": 0.031136974955910973,
+      "learning_rate": 4.569387226455776e-05,
+      "loss": 1.558,
+      "step": 2990
+    },
+    {
+      "epoch": 2.802007938360962,
+      "grad_norm": 0.032345263568839946,
+      "learning_rate": 4.5046711329508997e-05,
+      "loss": 1.5567,
+      "step": 3000
+    },
+    {
+      "epoch": 2.8113471865514827,
+      "grad_norm": 0.03075590840408306,
+      "learning_rate": 4.440283101896112e-05,
+      "loss": 1.5509,
+      "step": 3010
+    },
+    {
+      "epoch": 2.8206864347420035,
+      "grad_norm": 0.03212380955577142,
+      "learning_rate": 4.376226977155118e-05,
+      "loss": 1.5549,
+      "step": 3020
+    },
+    {
+      "epoch": 2.830025682932524,
+      "grad_norm": 0.030448994102361984,
+      "learning_rate": 4.3125065827773535e-05,
+      "loss": 1.5582,
+      "step": 3030
+    },
+    {
+      "epoch": 2.8393649311230447,
+      "grad_norm": 0.02952309346631166,
+      "learning_rate": 4.249125722769679e-05,
+      "loss": 1.5621,
+      "step": 3040
+    },
+    {
+      "epoch": 2.8487041793135655,
+      "grad_norm": 0.029801760584304334,
+      "learning_rate": 4.18608818086928e-05,
+      "loss": 1.5566,
+      "step": 3050
+    },
+    {
+      "epoch": 2.858043427504086,
+      "grad_norm": 0.029682850572275397,
+      "learning_rate": 4.12339772031781e-05,
+      "loss": 1.563,
+      "step": 3060
+    },
+    {
+      "epoch": 2.8673826756946066,
+      "grad_norm": 0.0295800536252474,
+      "learning_rate": 4.061058083636702e-05,
+      "loss": 1.5478,
+      "step": 3070
+    },
+    {
+      "epoch": 2.8767219238851274,
+      "grad_norm": 0.029683305961870896,
+      "learning_rate": 3.999072992403756e-05,
+      "loss": 1.5628,
+      "step": 3080
+    },
+    {
+      "epoch": 2.8860611720756477,
+      "grad_norm": 0.029644699672258072,
+      "learning_rate": 3.93744614703098e-05,
+      "loss": 1.5541,
+      "step": 3090
+    },
+    {
+      "epoch": 2.8954004202661685,
+      "grad_norm": 0.030518095362273494,
+      "learning_rate": 3.876181226543668e-05,
+      "loss": 1.5702,
+      "step": 3100
+    },
+    {
+      "epoch": 2.9047396684566893,
+      "grad_norm": 0.030536745251161155,
+      "learning_rate": 3.81528188836076e-05,
+      "loss": 1.5551,
+      "step": 3110
+    },
+    {
+      "epoch": 2.9140789166472096,
+      "grad_norm": 0.030847847626104093,
+      "learning_rate": 3.7547517680765244e-05,
+      "loss": 1.5595,
+      "step": 3120
+    },
+    {
+      "epoch": 2.9234181648377304,
+      "grad_norm": 0.029492648716320746,
+      "learning_rate": 3.6945944792434906e-05,
+      "loss": 1.5536,
+      "step": 3130
+    },
+    {
+      "epoch": 2.932757413028251,
+      "grad_norm": 0.03032565633382453,
+      "learning_rate": 3.634813613156753e-05,
+      "loss": 1.5545,
+      "step": 3140
+    },
+    {
+      "epoch": 2.942096661218772,
+      "grad_norm": 0.031152780923178743,
+      "learning_rate": 3.5754127386395496e-05,
+      "loss": 1.5553,
+      "step": 3150
+    },
+    {
+      "epoch": 2.951435909409293,
+      "grad_norm": 0.03107457614016718,
+      "learning_rate": 3.5163954018302313e-05,
+      "loss": 1.5612,
+      "step": 3160
+    },
+    {
+      "epoch": 2.960775157599813,
+      "grad_norm": 0.03104057970445527,
+      "learning_rate": 3.4577651259705545e-05,
+      "loss": 1.5461,
+      "step": 3170
+    },
+    {
+      "epoch": 2.970114405790334,
+      "grad_norm": 0.03135460216536421,
+      "learning_rate": 3.399525411195339e-05,
+      "loss": 1.556,
+      "step": 3180
+    },
+    {
+      "epoch": 2.9794536539808547,
+      "grad_norm": 0.03023549504443143,
+      "learning_rate": 3.3416797343235375e-05,
+      "loss": 1.5513,
+      "step": 3190
+    },
+    {
+      "epoch": 2.988792902171375,
+      "grad_norm": 0.030100263338637512,
+      "learning_rate": 3.284231548650649e-05,
+      "loss": 1.5646,
+      "step": 3200
+    },
+    {
+      "epoch": 2.998132150361896,
+      "grad_norm": 0.03034217414728692,
+      "learning_rate": 3.227184283742591e-05,
+      "loss": 1.5521,
+      "step": 3210
+    },
+    {
+      "epoch": 3.007588139154798,
+      "grad_norm": 0.06632657839957506,
+      "learning_rate": 3.17054134523093e-05,
+      "loss": 1.3861,
+      "step": 3220
+    },
+    {
+      "epoch": 3.016927387345319,
+      "grad_norm": 0.04494797332190053,
+      "learning_rate": 3.114306114609594e-05,
+      "loss": 1.2688,
+      "step": 3230
+    },
+    {
+      "epoch": 3.026266635535839,
+      "grad_norm": 0.04243454415323923,
+      "learning_rate": 3.058481949032998e-05,
+      "loss": 1.2568,
+      "step": 3240
+    },
+    {
+      "epoch": 3.03560588372636,
+      "grad_norm": 0.037189416438265135,
+      "learning_rate": 3.003072181115615e-05,
+      "loss": 1.2674,
+      "step": 3250
+    },
+    {
+      "epoch": 3.0449451319168808,
+      "grad_norm": 0.036342648540388975,
+      "learning_rate": 2.9480801187330308e-05,
+      "loss": 1.2553,
+      "step": 3260
+    },
+    {
+      "epoch": 3.0542843801074016,
+      "grad_norm": 0.03758108482902846,
+      "learning_rate": 2.8935090448244838e-05,
+      "loss": 1.2587,
+      "step": 3270
+    },
+    {
+      "epoch": 3.063623628297922,
+      "grad_norm": 0.03779156244551835,
+      "learning_rate": 2.8393622171968494e-05,
+      "loss": 1.2615,
+      "step": 3280
+    },
+    {
+      "epoch": 3.0729628764884427,
+      "grad_norm": 0.0354713302142897,
+      "learning_rate": 2.7856428683301827e-05,
+      "loss": 1.2552,
+      "step": 3290
+    },
+    {
+      "epoch": 3.0823021246789635,
+      "grad_norm": 0.03861733806219831,
+      "learning_rate": 2.7323542051847305e-05,
+      "loss": 1.2591,
+      "step": 3300
+    },
+    {
+      "epoch": 3.091641372869484,
+      "grad_norm": 0.03927047617762108,
+      "learning_rate": 2.6794994090094748e-05,
+      "loss": 1.2568,
+      "step": 3310
+    },
+    {
+      "epoch": 3.1009806210600046,
+      "grad_norm": 0.03904084831655504,
+      "learning_rate": 2.6270816351522352e-05,
+      "loss": 1.2548,
+      "step": 3320
+    },
+    {
+      "epoch": 3.1103198692505254,
+      "grad_norm": 0.037088038012932166,
+      "learning_rate": 2.57510401287128e-05,
+      "loss": 1.2535,
+      "step": 3330
+    },
+    {
+      "epoch": 3.119659117441046,
+      "grad_norm": 0.039689764947511294,
+      "learning_rate": 2.5235696451485334e-05,
+      "loss": 1.2527,
+      "step": 3340
+    },
+    {
+      "epoch": 3.1289983656315665,
+      "grad_norm": 0.03823754136616741,
+      "learning_rate": 2.4724816085043122e-05,
+      "loss": 1.2614,
+      "step": 3350
+    },
+    {
+      "epoch": 3.1383376138220873,
+      "grad_norm": 0.03869724476981111,
+      "learning_rate": 2.421842952813683e-05,
+      "loss": 1.2539,
+      "step": 3360
+    },
+    {
+      "epoch": 3.147676862012608,
+      "grad_norm": 0.03803608771547874,
+      "learning_rate": 2.371656701124376e-05,
+      "loss": 1.2491,
+      "step": 3370
+    },
+    {
+      "epoch": 3.1570161102031284,
+      "grad_norm": 0.03931544384220963,
+      "learning_rate": 2.321925849476314e-05,
+      "loss": 1.2505,
+      "step": 3380
+    },
+    {
+      "epoch": 3.1663553583936492,
+      "grad_norm": 0.039905322735836295,
+      "learning_rate": 2.2726533667227623e-05,
+      "loss": 1.2585,
+      "step": 3390
+    },
+    {
+      "epoch": 3.17569460658417,
+      "grad_norm": 0.03812720334866993,
+      "learning_rate": 2.223842194353084e-05,
+      "loss": 1.2589,
+      "step": 3400
+    },
+    {
+      "epoch": 3.185033854774691,
+      "grad_norm": 0.03970128613042995,
+      "learning_rate": 2.1754952463171385e-05,
+      "loss": 1.2547,
+      "step": 3410
+    },
+    {
+      "epoch": 3.194373102965211,
+      "grad_norm": 0.03968022448776454,
+      "learning_rate": 2.1276154088513322e-05,
+      "loss": 1.256,
+      "step": 3420
+    },
+    {
+      "epoch": 3.203712351155732,
+      "grad_norm": 0.038047431666726306,
+      "learning_rate": 2.080205540306307e-05,
+      "loss": 1.2618,
+      "step": 3430
+    },
+    {
+      "epoch": 3.2130515993462527,
+      "grad_norm": 0.03796892071652476,
+      "learning_rate": 2.0332684709763037e-05,
+      "loss": 1.258,
+      "step": 3440
+    },
+    {
+      "epoch": 3.222390847536773,
+      "grad_norm": 0.03743002539050354,
+      "learning_rate": 1.986807002930192e-05,
+      "loss": 1.2625,
+      "step": 3450
+    },
+    {
+      "epoch": 3.231730095727294,
+      "grad_norm": 0.03845080735099829,
+      "learning_rate": 1.940823909844198e-05,
+      "loss": 1.2583,
+      "step": 3460
+    },
+    {
+      "epoch": 3.2410693439178146,
+      "grad_norm": 0.0384612122659626,
+      "learning_rate": 1.895321936836324e-05,
+      "loss": 1.2504,
+      "step": 3470
+    },
+    {
+      "epoch": 3.2504085921083354,
+      "grad_norm": 0.03770473444960908,
+      "learning_rate": 1.8503038003024564e-05,
+      "loss": 1.2691,
+      "step": 3480
+    },
+    {
+      "epoch": 3.259747840298856,
+      "grad_norm": 0.0368880551504251,
+      "learning_rate": 1.8057721877542135e-05,
+      "loss": 1.2562,
+      "step": 3490
+    },
+    {
+      "epoch": 3.2690870884893766,
+      "grad_norm": 0.035849085577933516,
+      "learning_rate": 1.7617297576585024e-05,
+      "loss": 1.2557,
+      "step": 3500
+    },
+    {
+      "epoch": 3.2784263366798974,
+      "grad_norm": 0.03880537892593414,
+      "learning_rate": 1.7181791392788016e-05,
+      "loss": 1.2606,
+      "step": 3510
+    },
+    {
+      "epoch": 3.287765584870418,
+      "grad_norm": 0.036791883288228966,
+      "learning_rate": 1.6751229325182195e-05,
+      "loss": 1.2633,
+      "step": 3520
+    },
+    {
+      "epoch": 3.2971048330609385,
+      "grad_norm": 0.03888037048657713,
+      "learning_rate": 1.6325637077642597e-05,
+      "loss": 1.2517,
+      "step": 3530
+    },
+    {
+      "epoch": 3.3064440812514593,
+      "grad_norm": 0.03624411339308035,
+      "learning_rate": 1.590504005735396e-05,
+      "loss": 1.2601,
+      "step": 3540
+    },
+    {
+      "epoch": 3.31578332944198,
+      "grad_norm": 0.037212088295175624,
+      "learning_rate": 1.5489463373293766e-05,
+      "loss": 1.2572,
+      "step": 3550
+    },
+    {
+      "epoch": 3.3251225776325004,
+      "grad_norm": 0.037139053328387565,
+      "learning_rate": 1.5078931834733413e-05,
+      "loss": 1.2587,
+      "step": 3560
+    },
+    {
+      "epoch": 3.334461825823021,
+      "grad_norm": 0.038853212387254546,
+      "learning_rate": 1.467346994975708e-05,
+      "loss": 1.2598,
+      "step": 3570
+    },
+    {
+      "epoch": 3.343801074013542,
+      "grad_norm": 0.038068637710771305,
+      "learning_rate": 1.4273101923798593e-05,
+      "loss": 1.2511,
+      "step": 3580
+    },
+    {
+      "epoch": 3.3531403222040623,
+      "grad_norm": 0.03812184703208946,
+      "learning_rate": 1.3877851658196473e-05,
+      "loss": 1.264,
+      "step": 3590
+    },
+    {
+      "epoch": 3.362479570394583,
+      "grad_norm": 0.03544270078567587,
+      "learning_rate": 1.3487742748767052e-05,
+      "loss": 1.2566,
+      "step": 3600
+    },
+    {
+      "epoch": 3.371818818585104,
+      "grad_norm": 0.03729183894507396,
+      "learning_rate": 1.3102798484395761e-05,
+      "loss": 1.2554,
+      "step": 3610
+    },
+    {
+      "epoch": 3.3811580667756247,
+      "grad_norm": 0.03769688061241049,
+      "learning_rate": 1.2723041845646999e-05,
+      "loss": 1.2424,
+      "step": 3620
+    },
+    {
+      "epoch": 3.390497314966145,
+      "grad_norm": 0.03751980172858499,
+      "learning_rate": 1.2348495503391999e-05,
+      "loss": 1.2601,
+      "step": 3630
+    },
+    {
+      "epoch": 3.399836563156666,
+      "grad_norm": 0.03831727516561825,
+      "learning_rate": 1.1979181817455543e-05,
+      "loss": 1.2545,
+      "step": 3640
+    },
+    {
+      "epoch": 3.4091758113471866,
+      "grad_norm": 0.03929713475600568,
+      "learning_rate": 1.1615122835281156e-05,
+      "loss": 1.2608,
+      "step": 3650
+    },
+    {
+      "epoch": 3.4185150595377074,
+      "grad_norm": 0.04026553291819152,
+      "learning_rate": 1.1256340290614787e-05,
+      "loss": 1.2565,
+      "step": 3660
+    },
+    {
+      "epoch": 3.4278543077282277,
+      "grad_norm": 0.03669233459600138,
+      "learning_rate": 1.0902855602207451e-05,
+      "loss": 1.2512,
+      "step": 3670
+    },
+    {
+      "epoch": 3.4371935559187485,
+      "grad_norm": 0.036514113026692024,
+      "learning_rate": 1.0554689872536515e-05,
+      "loss": 1.2522,
+      "step": 3680
+    },
+    {
+      "epoch": 3.4465328041092693,
+      "grad_norm": 0.040574229602467475,
+      "learning_rate": 1.0211863886545859e-05,
+      "loss": 1.2653,
+      "step": 3690
+    },
+    {
+      "epoch": 3.4558720522997897,
+      "grad_norm": 0.037888410230506514,
+      "learning_rate": 9.87439811040518e-06,
+      "loss": 1.2553,
+      "step": 3700
+    },
+    {
+      "epoch": 3.4652113004903105,
+      "grad_norm": 0.037651161268116955,
+      "learning_rate": 9.542312690288036e-06,
+      "loss": 1.2513,
+      "step": 3710
+    },
+    {
+      "epoch": 3.4745505486808312,
+      "grad_norm": 0.03597355201152629,
+      "learning_rate": 9.215627451169317e-06,
+      "loss": 1.2545,
+      "step": 3720
+    },
+    {
+      "epoch": 3.483889796871352,
+      "grad_norm": 0.03806899793632852,
+      "learning_rate": 8.894361895641567e-06,
+      "loss": 1.2495,
+      "step": 3730
+    },
+    {
+      "epoch": 3.4932290450618724,
+      "grad_norm": 0.036576666432106365,
+      "learning_rate": 8.578535202750792e-06,
+      "loss": 1.258,
+      "step": 3740
+    },
+    {
+      "epoch": 3.502568293252393,
+      "grad_norm": 0.03719904682167482,
+      "learning_rate": 8.26816622685157e-06,
+      "loss": 1.2574,
+      "step": 3750
+    },
+    {
+      "epoch": 3.511907541442914,
+      "grad_norm": 0.035727663511753396,
+      "learning_rate": 7.963273496481294e-06,
+      "loss": 1.2605,
+      "step": 3760
+    },
+    {
+      "epoch": 3.5212467896334347,
+      "grad_norm": 0.036155845802186545,
+      "learning_rate": 7.663875213254246e-06,
+      "loss": 1.2607,
+      "step": 3770
+    },
+    {
+      "epoch": 3.530586037823955,
+      "grad_norm": 0.03686419389413766,
+      "learning_rate": 7.369989250774811e-06,
+      "loss": 1.2566,
+      "step": 3780
+    },
+    {
+      "epoch": 3.539925286014476,
+      "grad_norm": 0.03403684597854,
+      "learning_rate": 7.081633153570577e-06,
+      "loss": 1.2494,
+      "step": 3790
+    },
+    {
+      "epoch": 3.5492645342049967,
+      "grad_norm": 0.039208776340876904,
+      "learning_rate": 6.7988241360449126e-06,
+      "loss": 1.2578,
+      "step": 3800
+    },
+    {
+      "epoch": 3.558603782395517,
+      "grad_norm": 0.036990149740882036,
+      "learning_rate": 6.521579081449325e-06,
+      "loss": 1.2557,
+      "step": 3810
+    },
+    {
+      "epoch": 3.567943030586038,
+      "grad_norm": 0.037043700320972345,
+      "learning_rate": 6.249914540875446e-06,
+      "loss": 1.2466,
+      "step": 3820
+    },
+    {
+      "epoch": 3.5772822787765586,
+      "grad_norm": 0.035841902311675944,
+      "learning_rate": 5.983846732267118e-06,
+      "loss": 1.2562,
+      "step": 3830
+    },
+    {
+      "epoch": 3.586621526967079,
+      "grad_norm": 0.038201812926149784,
+      "learning_rate": 5.723391539452061e-06,
+      "loss": 1.2503,
+      "step": 3840
+    },
+    {
+      "epoch": 3.5959607751575997,
+      "grad_norm": 0.03786837415779077,
+      "learning_rate": 5.468564511193785e-06,
+      "loss": 1.2521,
+      "step": 3850
+    },
+    {
+      "epoch": 3.6053000233481205,
+      "grad_norm": 0.03734728180544562,
+      "learning_rate": 5.219380860263168e-06,
+      "loss": 1.254,
+      "step": 3860
+    },
+    {
+      "epoch": 3.6146392715386413,
+      "grad_norm": 0.034903789745343396,
+      "learning_rate": 4.975855462530465e-06,
+      "loss": 1.2496,
+      "step": 3870
+    },
+    {
+      "epoch": 3.623978519729162,
+      "grad_norm": 0.03765870695948473,
+      "learning_rate": 4.738002856077118e-06,
+      "loss": 1.2536,
+      "step": 3880
+    },
+    {
+      "epoch": 3.6333177679196824,
+      "grad_norm": 0.03513914326954358,
+      "learning_rate": 4.505837240327881e-06,
+      "loss": 1.2435,
+      "step": 3890
+    },
+    {
+      "epoch": 3.642657016110203,
+      "grad_norm": 0.03964268261724614,
+      "learning_rate": 4.279372475203181e-06,
+      "loss": 1.2593,
+      "step": 3900
+    },
+    {
+      "epoch": 3.651996264300724,
+      "grad_norm": 0.03469425212172376,
+      "learning_rate": 4.058622080291652e-06,
+      "loss": 1.2532,
+      "step": 3910
+    },
+    {
+      "epoch": 3.6613355124912443,
+      "grad_norm": 0.03598049947724251,
+      "learning_rate": 3.843599234043038e-06,
+      "loss": 1.2556,
+      "step": 3920
+    },
+    {
+      "epoch": 3.670674760681765,
+      "grad_norm": 0.03651926246151059,
+      "learning_rate": 3.6343167729815163e-06,
+      "loss": 1.2587,
+      "step": 3930
+    },
+    {
+      "epoch": 3.680014008872286,
+      "grad_norm": 0.03552581291333623,
+      "learning_rate": 3.430787190939322e-06,
+      "loss": 1.2464,
+      "step": 3940
+    },
+    {
+      "epoch": 3.6893532570628063,
+      "grad_norm": 0.037931227770931816,
+      "learning_rate": 3.233022638310901e-06,
+      "loss": 1.2552,
+      "step": 3950
+    },
+    {
+      "epoch": 3.698692505253327,
+      "grad_norm": 0.036304173322246995,
+      "learning_rate": 3.0410349213275567e-06,
+      "loss": 1.2482,
+      "step": 3960
+    },
+    {
+      "epoch": 3.708031753443848,
+      "grad_norm": 0.03538930810416937,
+      "learning_rate": 2.8548355013526153e-06,
+      "loss": 1.2595,
+      "step": 3970
+    },
+    {
+      "epoch": 3.717371001634368,
+      "grad_norm": 0.03665091668672923,
+      "learning_rate": 2.6744354941972473e-06,
+      "loss": 1.2572,
+      "step": 3980
+    },
+    {
+      "epoch": 3.726710249824889,
+      "grad_norm": 0.03652781100835373,
+      "learning_rate": 2.4998456694568016e-06,
+      "loss": 1.2605,
+      "step": 3990
+    },
+    {
+      "epoch": 3.7360494980154098,
+      "grad_norm": 0.03552338730654514,
+      "learning_rate": 2.331076449867975e-06,
+      "loss": 1.2491,
+      "step": 4000
+    },
+    {
+      "epoch": 3.7453887462059305,
+      "grad_norm": 0.036787806397377276,
+      "learning_rate": 2.168137910686485e-06,
+      "loss": 1.2554,
+      "step": 4010
+    },
+    {
+      "epoch": 3.7547279943964513,
+      "grad_norm": 0.03691275737074826,
+      "learning_rate": 2.011039779085655e-06,
+      "loss": 1.2489,
+      "step": 4020
+    },
+    {
+      "epoch": 3.7640672425869717,
+      "grad_norm": 0.03543531097242808,
+      "learning_rate": 1.8597914335757083e-06,
+      "loss": 1.2541,
+      "step": 4030
+    },
+    {
+      "epoch": 3.7734064907774925,
+      "grad_norm": 0.03588804476631641,
+      "learning_rate": 1.7144019034438851e-06,
+      "loss": 1.2567,
+      "step": 4040
+    },
+    {
+      "epoch": 3.7827457389680132,
+      "grad_norm": 0.036333508073258726,
+      "learning_rate": 1.5748798682154176e-06,
+      "loss": 1.2469,
+      "step": 4050
+    },
+    {
+      "epoch": 3.7920849871585336,
+      "grad_norm": 0.03671560487120199,
+      "learning_rate": 1.4412336571353102e-06,
+      "loss": 1.2549,
+      "step": 4060
+    },
+    {
+      "epoch": 3.8014242353490544,
+      "grad_norm": 0.03385836988167257,
+      "learning_rate": 1.3134712486712163e-06,
+      "loss": 1.2472,
+      "step": 4070
+    },
+    {
+      "epoch": 3.810763483539575,
+      "grad_norm": 0.03514099866475247,
+      "learning_rate": 1.191600270037041e-06,
+      "loss": 1.25,
+      "step": 4080
+    },
+    {
+      "epoch": 3.8201027317300955,
+      "grad_norm": 0.036579757858228146,
+      "learning_rate": 1.075627996737627e-06,
+      "loss": 1.2554,
+      "step": 4090
+    },
+    {
+      "epoch": 3.8294419799206163,
+      "grad_norm": 0.035474764761714114,
+      "learning_rate": 9.655613521344365e-07,
+      "loss": 1.2493,
+      "step": 4100
+    },
+    {
+      "epoch": 3.838781228111137,
+      "grad_norm": 0.035780463770959316,
+      "learning_rate": 8.614069070322473e-07,
+      "loss": 1.2516,
+      "step": 4110
+    },
+    {
+      "epoch": 3.848120476301658,
+      "grad_norm": 0.03667535890217496,
+      "learning_rate": 7.631708792868453e-07,
+      "loss": 1.2525,
+      "step": 4120
+    },
+    {
+      "epoch": 3.857459724492178,
+      "grad_norm": 0.03866821184004067,
+      "learning_rate": 6.708591334338654e-07,
+      "loss": 1.2437,
+      "step": 4130
+    },
+    {
+      "epoch": 3.866798972682699,
+      "grad_norm": 0.035931245102736833,
+      "learning_rate": 5.844771803386828e-07,
+      "loss": 1.2516,
+      "step": 4140
+    },
+    {
+      "epoch": 3.87613822087322,
+      "grad_norm": 0.034107023911041444,
+      "learning_rate": 5.040301768673761e-07,
+      "loss": 1.2472,
+      "step": 4150
+    },
+    {
+      "epoch": 3.8854774690637406,
+      "grad_norm": 0.03555842514434786,
+      "learning_rate": 4.2952292557896234e-07,
+      "loss": 1.2454,
+      "step": 4160
+    },
+    {
+      "epoch": 3.894816717254261,
+      "grad_norm": 0.034240935833829624,
+      "learning_rate": 3.6095987443860445e-07,
+      "loss": 1.2528,
+      "step": 4170
+    },
+    {
+      "epoch": 3.9041559654447817,
+      "grad_norm": 0.03666542915960847,
+      "learning_rate": 2.9834511655211226e-07,
+      "loss": 1.2519,
+      "step": 4180
+    },
+    {
+      "epoch": 3.9134952136353025,
+      "grad_norm": 0.037655412271416625,
+      "learning_rate": 2.4168238992160474e-07,
+      "loss": 1.2555,
+      "step": 4190
+    },
+    {
+      "epoch": 3.922834461825823,
+      "grad_norm": 0.036197022796120186,
+      "learning_rate": 1.9097507722231067e-07,
+      "loss": 1.2591,
+      "step": 4200
+    },
+    {
+      "epoch": 3.9321737100163436,
+      "grad_norm": 0.036305484542715774,
+      "learning_rate": 1.462262056006969e-07,
+      "loss": 1.2474,
+      "step": 4210
+    },
+    {
+      "epoch": 3.9415129582068644,
+      "grad_norm": 0.036303954460438236,
+      "learning_rate": 1.074384464936684e-07,
+      "loss": 1.2418,
+      "step": 4220
+    },
+    {
+      "epoch": 3.9508522063973848,
+      "grad_norm": 0.03569049340659871,
+      "learning_rate": 7.461411546916263e-08,
+      "loss": 1.2496,
+      "step": 4230
+    },
+    {
+      "epoch": 3.9601914545879056,
+      "grad_norm": 0.03796994227856852,
+      "learning_rate": 4.775517208788216e-08,
+      "loss": 1.2479,
+      "step": 4240
+    },
+    {
+      "epoch": 3.9695307027784263,
+      "grad_norm": 0.03394281504926072,
+      "learning_rate": 2.6863219786299465e-08,
+      "loss": 1.2456,
+      "step": 4250
+    },
+    {
+      "epoch": 3.978869950968947,
+      "grad_norm": 0.035760734149513904,
+      "learning_rate": 1.1939505780966808e-08,
+      "loss": 1.2542,
+      "step": 4260
+    },
+    {
+      "epoch": 3.988209199159468,
+      "grad_norm": 0.036716516315121565,
+      "learning_rate": 2.9849209940091547e-09,
+      "loss": 1.2625,
+      "step": 4270
+    },
+    {
+      "epoch": 3.9975484473499883,
+      "grad_norm": 0.03584256618365922,
+      "learning_rate": 0.0,
+      "loss": 1.2575,
+      "step": 4280
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 4280,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 100,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.465521953285163e+20,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

l2-13b-ga/checkpoint-4280/zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

l2-13b-ga/checkpoint-700/config.json

@@ -0,0 +1,30 @@
+{
+  "_name_or_path": "meta-llama/Llama-2-13b-hf",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 5120,
+  "initializer_range": 0.02,
+  "intermediate_size": 13824,
+  "max_position_embeddings": 4096,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "num_attention_heads": 40,
+  "num_hidden_layers": 40,
+  "num_key_value_heads": 40,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.46.3",
+  "use_cache": true,
+  "vocab_size": 35483
+}

l2-13b-ga/checkpoint-700/generation_config.json

@@ -0,0 +1,10 @@
+{
+  "bos_token_id": 1,
+  "do_sample": true,
+  "eos_token_id": 2,
+  "max_length": 4096,
+  "pad_token_id": 0,
+  "temperature": 0.6,
+  "top_p": 0.9,
+  "transformers_version": "4.46.3"
+}

l2-13b-ga/checkpoint-700/latest

@@ -0,0 +1 @@
+global_step700

l2-13b-ga/checkpoint-700/model.safetensors.index.json

@@ -0,0 +1,370 @@
+{
+  "metadata": {
+    "total_size": 26103060480
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00006-of-00006.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.10.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.11.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.12.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.13.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.14.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.15.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.15.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.16.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.17.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.18.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.19.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.20.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.down_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.21.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.mlp.gate_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.mlp.up_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.22.self_attn.k_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.q_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.22.self_attn.v_proj.weight": "model-00003-of-00006.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.23.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.24.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.25.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.26.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.27.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.28.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.input_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.down_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.mlp.up_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.post_attention_layernorm.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.29.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.30.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.mlp.gate_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.30.self_attn.k_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.o_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.q_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.30.self_attn.v_proj.weight": "model-00004-of-00006.safetensors",
+    "model.layers.31.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.31.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.32.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.33.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.34.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.35.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.36.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.input_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.down_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.gate_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.mlp.up_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.post_attention_layernorm.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.37.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.input_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.down_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.gate_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.mlp.up_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.post_attention_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.38.self_attn.k_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.o_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.q_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.38.self_attn.v_proj.weight": "model-00005-of-00006.safetensors",
+    "model.layers.39.input_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.down_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.gate_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.mlp.up_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.post_attention_layernorm.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.k_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.o_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.q_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.39.self_attn.v_proj.weight": "model-00006-of-00006.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.5.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.down_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.gate_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.mlp.up_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.6.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.self_attn.k_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.7.self_attn.q_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.7.self_attn.v_proj.weight": "model-00001-of-00006.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.8.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.down_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.gate_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.mlp.up_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.k_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.q_proj.weight": "model-00002-of-00006.safetensors",
+    "model.layers.9.self_attn.v_proj.weight": "model-00002-of-00006.safetensors",
+    "model.norm.weight": "model-00006-of-00006.safetensors"
+  }
+}

l2-13b-ga/checkpoint-700/special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

l2-13b-ga/checkpoint-700/tokenizer_config.json

@@ -0,0 +1,42 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": true,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": true,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

l2-13b-ga/checkpoint-700/trainer_state.json

@@ -0,0 +1,530 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.6537473733364464,
+  "eval_steps": 500,
+  "global_step": 700,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.0009339248190520663,
+      "grad_norm": 6.638877692627699,
+      "learning_rate": 9.345794392523364e-07,
+      "loss": 9.2917,
+      "step": 1
+    },
+    {
+      "epoch": 0.009339248190520663,
+      "grad_norm": 1.1560921335705272,
+      "learning_rate": 9.345794392523365e-06,
+      "loss": 9.0876,
+      "step": 10
+    },
+    {
+      "epoch": 0.018678496381041326,
+      "grad_norm": 0.8415132296956432,
+      "learning_rate": 1.869158878504673e-05,
+      "loss": 8.2164,
+      "step": 20
+    },
+    {
+      "epoch": 0.02801774457156199,
+      "grad_norm": 0.45381630992958155,
+      "learning_rate": 2.8037383177570094e-05,
+      "loss": 7.5184,
+      "step": 30
+    },
+    {
+      "epoch": 0.03735699276208265,
+      "grad_norm": 0.8400636107958425,
+      "learning_rate": 3.738317757009346e-05,
+      "loss": 6.6507,
+      "step": 40
+    },
+    {
+      "epoch": 0.046696240952603316,
+      "grad_norm": 0.557696240829066,
+      "learning_rate": 4.672897196261683e-05,
+      "loss": 5.8909,
+      "step": 50
+    },
+    {
+      "epoch": 0.05603548914312398,
+      "grad_norm": 0.3971996057467842,
+      "learning_rate": 5.607476635514019e-05,
+      "loss": 5.4127,
+      "step": 60
+    },
+    {
+      "epoch": 0.06537473733364464,
+      "grad_norm": 0.2932710540265688,
+      "learning_rate": 6.542056074766355e-05,
+      "loss": 5.0106,
+      "step": 70
+    },
+    {
+      "epoch": 0.0747139855241653,
+      "grad_norm": 0.3682690443551033,
+      "learning_rate": 7.476635514018692e-05,
+      "loss": 4.6042,
+      "step": 80
+    },
+    {
+      "epoch": 0.08405323371468597,
+      "grad_norm": 0.3132971920011515,
+      "learning_rate": 8.411214953271028e-05,
+      "loss": 4.2031,
+      "step": 90
+    },
+    {
+      "epoch": 0.09339248190520663,
+      "grad_norm": 0.6731868159213446,
+      "learning_rate": 9.345794392523365e-05,
+      "loss": 3.9423,
+      "step": 100
+    },
+    {
+      "epoch": 0.1027317300957273,
+      "grad_norm": 0.27848867836763197,
+      "learning_rate": 0.000102803738317757,
+      "loss": 3.7157,
+      "step": 110
+    },
+    {
+      "epoch": 0.11207097828624796,
+      "grad_norm": 0.24642109032991807,
+      "learning_rate": 0.00011214953271028037,
+      "loss": 3.4516,
+      "step": 120
+    },
+    {
+      "epoch": 0.12141022647676862,
+      "grad_norm": 0.25717384664029797,
+      "learning_rate": 0.00012149532710280373,
+      "loss": 3.2167,
+      "step": 130
+    },
+    {
+      "epoch": 0.13074947466728928,
+      "grad_norm": 0.20912922668565637,
+      "learning_rate": 0.0001308411214953271,
+      "loss": 3.0237,
+      "step": 140
+    },
+    {
+      "epoch": 0.14008872285780993,
+      "grad_norm": 0.15805888388706113,
+      "learning_rate": 0.00014018691588785047,
+      "loss": 2.8529,
+      "step": 150
+    },
+    {
+      "epoch": 0.1494279710483306,
+      "grad_norm": 0.23370349497479534,
+      "learning_rate": 0.00014953271028037384,
+      "loss": 2.7078,
+      "step": 160
+    },
+    {
+      "epoch": 0.15876721923885126,
+      "grad_norm": 0.1802138633012483,
+      "learning_rate": 0.0001588785046728972,
+      "loss": 2.6115,
+      "step": 170
+    },
+    {
+      "epoch": 0.16810646742937194,
+      "grad_norm": 0.13354347610039718,
+      "learning_rate": 0.00016822429906542056,
+      "loss": 2.5309,
+      "step": 180
+    },
+    {
+      "epoch": 0.17744571561989259,
+      "grad_norm": 0.09414865188086892,
+      "learning_rate": 0.00017757009345794393,
+      "loss": 2.4452,
+      "step": 190
+    },
+    {
+      "epoch": 0.18678496381041326,
+      "grad_norm": 0.08333601554768896,
+      "learning_rate": 0.0001869158878504673,
+      "loss": 2.3832,
+      "step": 200
+    },
+    {
+      "epoch": 0.1961242120009339,
+      "grad_norm": 0.15926414699806835,
+      "learning_rate": 0.00019626168224299065,
+      "loss": 2.3492,
+      "step": 210
+    },
+    {
+      "epoch": 0.2054634601914546,
+      "grad_norm": 0.09492820761057012,
+      "learning_rate": 0.0001999989254250208,
+      "loss": 2.323,
+      "step": 220
+    },
+    {
+      "epoch": 0.21480270838197524,
+      "grad_norm": 0.0801349259356147,
+      "learning_rate": 0.00019999235866155886,
+      "loss": 2.2731,
+      "step": 230
+    },
+    {
+      "epoch": 0.22414195657249592,
+      "grad_norm": 0.12210960524693895,
+      "learning_rate": 0.00019997982251228469,
+      "loss": 2.2433,
+      "step": 240
+    },
+    {
+      "epoch": 0.23348120476301656,
+      "grad_norm": 3.14289498732125,
+      "learning_rate": 0.00019996131772558666,
+      "loss": 3.2769,
+      "step": 250
+    },
+    {
+      "epoch": 0.24282045295353724,
+      "grad_norm": 1.632940983166179,
+      "learning_rate": 0.00019993684540617132,
+      "loss": 4.9343,
+      "step": 260
+    },
+    {
+      "epoch": 0.2521597011440579,
+      "grad_norm": 3.4831252230225416,
+      "learning_rate": 0.00019990640701499736,
+      "loss": 4.2768,
+      "step": 270
+    },
+    {
+      "epoch": 0.26149894933457857,
+      "grad_norm": 1.6069045920523788,
+      "learning_rate": 0.00019987000436918874,
+      "loss": 5.9581,
+      "step": 280
+    },
+    {
+      "epoch": 0.27083819752509924,
+      "grad_norm": 0.2220907936615993,
+      "learning_rate": 0.00019982763964192585,
+      "loss": 3.8228,
+      "step": 290
+    },
+    {
+      "epoch": 0.28017744571561987,
+      "grad_norm": 0.24737284913291765,
+      "learning_rate": 0.00019977931536231596,
+      "loss": 3.1413,
+      "step": 300
+    },
+    {
+      "epoch": 0.28951669390614054,
+      "grad_norm": 4.010404518241152,
+      "learning_rate": 0.00019972503441524224,
+      "loss": 2.8432,
+      "step": 310
+    },
+    {
+      "epoch": 0.2988559420966612,
+      "grad_norm": 0.1515583580811596,
+      "learning_rate": 0.00019966480004119142,
+      "loss": 2.7859,
+      "step": 320
+    },
+    {
+      "epoch": 0.3081951902871819,
+      "grad_norm": 0.11259395750650594,
+      "learning_rate": 0.00019959861583606045,
+      "loss": 2.5821,
+      "step": 330
+    },
+    {
+      "epoch": 0.3175344384777025,
+      "grad_norm": 0.22514797814956813,
+      "learning_rate": 0.00019952648575094183,
+      "loss": 2.4517,
+      "step": 340
+    },
+    {
+      "epoch": 0.3268736866682232,
+      "grad_norm": 0.08040136172033542,
+      "learning_rate": 0.00019944841409188767,
+      "loss": 2.3794,
+      "step": 350
+    },
+    {
+      "epoch": 0.3362129348587439,
+      "grad_norm": 0.054758073593565354,
+      "learning_rate": 0.00019936440551965263,
+      "loss": 2.3232,
+      "step": 360
+    },
+    {
+      "epoch": 0.34555218304926455,
+      "grad_norm": 0.06742998909645591,
+      "learning_rate": 0.00019927446504941577,
+      "loss": 2.2776,
+      "step": 370
+    },
+    {
+      "epoch": 0.35489143123978517,
+      "grad_norm": 0.048780907584876736,
+      "learning_rate": 0.00019917859805048096,
+      "loss": 2.2376,
+      "step": 380
+    },
+    {
+      "epoch": 0.36423067943030585,
+      "grad_norm": 0.0475325963052214,
+      "learning_rate": 0.00019907681024595663,
+      "loss": 2.2191,
+      "step": 390
+    },
+    {
+      "epoch": 0.3735699276208265,
+      "grad_norm": 0.054089563211590065,
+      "learning_rate": 0.00019896910771241387,
+      "loss": 2.1961,
+      "step": 400
+    },
+    {
+      "epoch": 0.3829091758113472,
+      "grad_norm": 0.21798406131864823,
+      "learning_rate": 0.00019885549687952372,
+      "loss": 2.2078,
+      "step": 410
+    },
+    {
+      "epoch": 0.3922484240018678,
+      "grad_norm": 0.8673185709111124,
+      "learning_rate": 0.00019873598452967338,
+      "loss": 2.3731,
+      "step": 420
+    },
+    {
+      "epoch": 0.4015876721923885,
+      "grad_norm": 0.22424350669971718,
+      "learning_rate": 0.0001986105777975613,
+      "loss": 2.6195,
+      "step": 430
+    },
+    {
+      "epoch": 0.4109269203829092,
+      "grad_norm": 0.307418135168262,
+      "learning_rate": 0.00019847928416977126,
+      "loss": 2.3624,
+      "step": 440
+    },
+    {
+      "epoch": 0.42026616857342985,
+      "grad_norm": 0.07944722668080402,
+      "learning_rate": 0.00019834211148432536,
+      "loss": 2.2799,
+      "step": 450
+    },
+    {
+      "epoch": 0.4296054167639505,
+      "grad_norm": 0.18146933758664588,
+      "learning_rate": 0.00019819906793021614,
+      "loss": 2.2177,
+      "step": 460
+    },
+    {
+      "epoch": 0.43894466495447115,
+      "grad_norm": 0.07035825837333018,
+      "learning_rate": 0.0001980501620469178,
+      "loss": 2.1767,
+      "step": 470
+    },
+    {
+      "epoch": 0.44828391314499183,
+      "grad_norm": 0.04596186944454228,
+      "learning_rate": 0.0001978954027238763,
+      "loss": 2.1598,
+      "step": 480
+    },
+    {
+      "epoch": 0.4576231613355125,
+      "grad_norm": 0.041342347745088055,
+      "learning_rate": 0.0001977347991999786,
+      "loss": 2.131,
+      "step": 490
+    },
+    {
+      "epoch": 0.46696240952603313,
+      "grad_norm": 0.04172063219841485,
+      "learning_rate": 0.00019756836106300137,
+      "loss": 2.1231,
+      "step": 500
+    },
+    {
+      "epoch": 0.4763016577165538,
+      "grad_norm": 0.03373646457711144,
+      "learning_rate": 0.00019739609824903843,
+      "loss": 2.1146,
+      "step": 510
+    },
+    {
+      "epoch": 0.4856409059070745,
+      "grad_norm": 0.03736871030676605,
+      "learning_rate": 0.00019721802104190748,
+      "loss": 2.1003,
+      "step": 520
+    },
+    {
+      "epoch": 0.49498015409759516,
+      "grad_norm": 0.033931028038211034,
+      "learning_rate": 0.00019703414007253645,
+      "loss": 2.0983,
+      "step": 530
+    },
+    {
+      "epoch": 0.5043194022881158,
+      "grad_norm": 0.03790055446070549,
+      "learning_rate": 0.00019684446631832868,
+      "loss": 2.092,
+      "step": 540
+    },
+    {
+      "epoch": 0.5136586504786365,
+      "grad_norm": 0.030956192803893078,
+      "learning_rate": 0.00019664901110250758,
+      "loss": 2.0807,
+      "step": 550
+    },
+    {
+      "epoch": 0.5229978986691571,
+      "grad_norm": 0.03542530209935129,
+      "learning_rate": 0.00019644778609344068,
+      "loss": 2.0773,
+      "step": 560
+    },
+    {
+      "epoch": 0.5323371468596778,
+      "grad_norm": 0.040947757568902336,
+      "learning_rate": 0.00019624080330394306,
+      "loss": 2.0649,
+      "step": 570
+    },
+    {
+      "epoch": 0.5416763950501985,
+      "grad_norm": 0.034273415973688146,
+      "learning_rate": 0.00019602807509056018,
+      "loss": 2.0479,
+      "step": 580
+    },
+    {
+      "epoch": 0.5510156432407192,
+      "grad_norm": 0.031427481498873144,
+      "learning_rate": 0.00019580961415283028,
+      "loss": 2.0563,
+      "step": 590
+    },
+    {
+      "epoch": 0.5603548914312397,
+      "grad_norm": 0.03141549752041532,
+      "learning_rate": 0.00019558543353252611,
+      "loss": 2.0503,
+      "step": 600
+    },
+    {
+      "epoch": 0.5696941396217604,
+      "grad_norm": 0.033012392726428204,
+      "learning_rate": 0.00019535554661287652,
+      "loss": 2.0389,
+      "step": 610
+    },
+    {
+      "epoch": 0.5790333878122811,
+      "grad_norm": 0.02913261992661444,
+      "learning_rate": 0.0001951199671177673,
+      "loss": 2.036,
+      "step": 620
+    },
+    {
+      "epoch": 0.5883726360028018,
+      "grad_norm": 0.030543903708435332,
+      "learning_rate": 0.00019487870911092214,
+      "loss": 2.0326,
+      "step": 630
+    },
+    {
+      "epoch": 0.5977118841933224,
+      "grad_norm": 0.03215005545393897,
+      "learning_rate": 0.00019463178699506277,
+      "loss": 2.0231,
+      "step": 640
+    },
+    {
+      "epoch": 0.6070511323838431,
+      "grad_norm": 0.03823630791937631,
+      "learning_rate": 0.00019437921551104933,
+      "loss": 2.0293,
+      "step": 650
+    },
+    {
+      "epoch": 0.6163903805743638,
+      "grad_norm": 0.03200103149471209,
+      "learning_rate": 0.00019412100973700038,
+      "loss": 2.017,
+      "step": 660
+    },
+    {
+      "epoch": 0.6257296287648845,
+      "grad_norm": 0.03841804153577787,
+      "learning_rate": 0.00019385718508739262,
+      "loss": 2.0135,
+      "step": 670
+    },
+    {
+      "epoch": 0.635068876955405,
+      "grad_norm": 0.03052396655271533,
+      "learning_rate": 0.0001935877573121407,
+      "loss": 2.0237,
+      "step": 680
+    },
+    {
+      "epoch": 0.6444081251459257,
+      "grad_norm": 0.033817837533771815,
+      "learning_rate": 0.00019331274249565717,
+      "loss": 2.0069,
+      "step": 690
+    },
+    {
+      "epoch": 0.6537473733364464,
+      "grad_norm": 0.028286999650643876,
+      "learning_rate": 0.00019303215705589194,
+      "loss": 2.0112,
+      "step": 700
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 4280,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 4,
+  "save_steps": 100,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 5.667909740273861e+19,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

l2-13b-ga/checkpoint-700/zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)