richardyoung/dolphin-2.9.1-yi-1.5-34b-heretic

This is a decensored version of dphn/dolphin-2.9.1-yi-1.5-34b, made using Heretic v1.0.1

Abliteration parameters

Parameter	Value
direction_index	42.46
attn.o_proj.max_weight	1.43
attn.o_proj.max_weight_position	41.27
attn.o_proj.min_weight	1.05
attn.o_proj.min_weight_distance	26.25
mlp.down_proj.max_weight	1.19
mlp.down_proj.max_weight_position	52.68
mlp.down_proj.min_weight	0.51
mlp.down_proj.min_weight_distance	5.54

Performance

Metric	This model	Original model (dphn/dolphin-2.9.1-yi-1.5-34b)
KL divergence	0.06	0 (by definition)
Refusals	11/100	40/100

---

Dolphin 2.9.1 Yi 1.5 34b 🐬

Curated and trained by Eric Hartford, Lucas Atkins, and Fernando Fernandes, and Cognitive Computations

This is our most spectacular outcome ever. FFT, all parameters, 16bit. 77.4 MMLU on 34b. And it talks like a dream.

Although the max positional embeddings is 4k, we used rope theta of 1000000.0 and we trained with sequence length 8k. We plan to train on the upcoming 32k version as well.

Website: https://dphn.ai
Twitter: https://x.com/dphnAI
Web Chat: https://chat.dphn.ai
Telegram bot: https://t.me/DolphinAI_bot

Our appreciation for the sponsors of Dolphin 2.9.1:

• Crusoe Cloud - provided excellent on-demand 8xH100 node
• OnDemand - provided inference sponsorship

This model is based on Yi-1.5-34b, and is governed by apache 2.0 license.

The base model has 4k context, but we used rope theta of 1000000.0 and the full-weight fine-tuning was with 8k sequence length.

Dolphin 2.9.1 uses ChatML prompt template format.

example:

<|im_start|>system
You are Dolphin, a helpful AI assistant.<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

Dolphin-2.9.1 has a variety of instruction, conversational, and coding skills. It also has initial agentic abilities and supports function calling.

Dolphin is uncensored. We have filtered the dataset to remove alignment and bias. This makes the model more compliant. You are advised to implement your own alignment layer before exposing the model as a service. It will be highly compliant with any requests, even unethical ones. Please read my blog post about uncensored models. https://erichartford.com/uncensored-models You are responsible for any content you create using this model. Enjoy responsibly.

Dolphin is licensed according to apache 2.0 license. We grant permission for any use, including commercial. Dolphin was trained on data generated from GPT4, among other models.

Evals

Training

See axolotl config

axolotl version: 0.4.0

base_model: 01-ai/Yi-1.5-34B
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer
trust_remote_code: true

# load_in_8bit: false
# load_in_4bit: true
# strict: false

# adapter: qlora
# lora_modules_to_save: [embed_tokens, lm_head]

# lora_r: 32
# lora_alpha: 16
# lora_dropout: 0.05
# lora_target_linear: True
# lora_fan_in_fan_out:

datasets:
  - path: /workspace/datasets/dolphin-2.9/dolphin201-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/dolphin-coder-translate-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/dolphin-coder-codegen-sharegpt2.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/m-a-p_Code-Feedback-sharegpt-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/m-a-p_CodeFeedback-Filtered-Instruction-sharegpt-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/not_samantha_norefusals.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/Orca-Math-resort-unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/agent_instruct_react_unfiltered.jsonl
    type: sharegpt  
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_instruct_j1s1_3k_unfiltered.jsonl
    type: sharegpt  
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_negative_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_react_10p_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/toolbench_tflan_cot_30p_unfiltered.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/dolphin-2.9/openhermes200k_unfiltered.jsonl
    type: sharegpt 
    conversation: chatml

chat_template: chatml

dataset_prepared_path: yi34b
val_set_size: 0.01
output_dir: ./out-yi

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

wandb_project: dolphin-2.9-yi-34b
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 3
optimizer: adamw_8bit
lr_scheduler: cosine
learning_rate: 1e-5

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: true

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
# resume_from_checkpoint: /workspace/axolotl/dbrx-checkpoint
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
saves_per_epoch: 4
save_total_limit: 2
save_steps:
debug:
deepspeed: /workspace/axolotl/deepspeed_configs/zero3_bf16.json
weight_decay: 0.05
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<|startoftext|>"
  eos_token: "<|im_end|>"
  pad_token: "<unk>"
  unk_token: "<unk>"
tokens:
  - "<|im_start|>"
  

out-yi

This model is a fine-tuned version of 01-ai/Yi-1.5-34B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4425

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
0.6265	0.0	1	0.6035
0.4674	0.25	327	0.4344
0.4337	0.5	654	0.4250
0.4346	0.75	981	0.4179
0.3985	1.0	1308	0.4118
0.3128	1.23	1635	0.4201
0.3261	1.48	1962	0.4157
0.3259	1.73	2289	0.4122
0.3126	1.98	2616	0.4079
0.2265	2.21	2943	0.4441
0.2297	2.46	3270	0.4427
0.2424	2.71	3597	0.4425

Framework versions

• Transformers 4.40.0.dev0
• Pytorch 2.2.2+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0

---

Important Disclaimer

This model has been modified to remove safety guardrails and refusal behaviors.

Intended Use

• Research and educational purposes
• Understanding model behavior and limitations
• Creative writing and roleplay with consenting adults
• Red-teaming and safety research

Not Intended For

• Generating harmful, illegal, or unethical content
• Harassment, abuse, or malicious activities
• Misinformation or deception
• Any use that violates applicable laws

User Responsibility

By using this model, you acknowledge that:

1. You are solely responsible for how you use this model and any content it generates
2. The model creator accepts no liability for misuse or harmful outputs
3. You will comply with all applicable laws and ethical guidelines
4. You understand this model may produce inaccurate, biased, or inappropriate content

Technical Note

This model was created using abliteration techniques that suppress the "refusal direction" in the model's activation space. This does not add new capabilities—it only removes trained refusal behaviors from the base model.

Use responsibly. You have been warned.

---