Upload folder using huggingface_hub
Browse files- adapter_config.json +2 -2
- adapter_model.safetensors +2 -2
- optimizer.pt +2 -2
- rng_state.pth +1 -1
- scheduler.pt +1 -1
- trainer_state.json +793 -443
- training_args.bin +1 -1
adapter_config.json
CHANGED
|
@@ -12,14 +12,14 @@
|
|
| 12 |
"layers_pattern": null,
|
| 13 |
"layers_to_transform": null,
|
| 14 |
"loftq_config": {},
|
| 15 |
-
"lora_alpha":
|
| 16 |
"lora_bias": false,
|
| 17 |
"lora_dropout": 0,
|
| 18 |
"megatron_config": null,
|
| 19 |
"megatron_core": "megatron.core",
|
| 20 |
"modules_to_save": null,
|
| 21 |
"peft_type": "LORA",
|
| 22 |
-
"r":
|
| 23 |
"rank_pattern": {},
|
| 24 |
"revision": null,
|
| 25 |
"target_modules": [
|
|
|
|
| 12 |
"layers_pattern": null,
|
| 13 |
"layers_to_transform": null,
|
| 14 |
"loftq_config": {},
|
| 15 |
+
"lora_alpha": 112,
|
| 16 |
"lora_bias": false,
|
| 17 |
"lora_dropout": 0,
|
| 18 |
"megatron_config": null,
|
| 19 |
"megatron_core": "megatron.core",
|
| 20 |
"modules_to_save": null,
|
| 21 |
"peft_type": "LORA",
|
| 22 |
+
"r": 56,
|
| 23 |
"rank_pattern": {},
|
| 24 |
"revision": null,
|
| 25 |
"target_modules": [
|
adapter_model.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1767407cde1f093e295836dca6f0a8fe3280307f21de2f9ac5a87c96ac476e21
|
| 3 |
+
size 161515608
|
optimizer.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:1b03158d3c32281c45bae11494452aff7910950a34011e853f3d6c1c18d8651b
|
| 3 |
+
size 323181259
|
rng_state.pth
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 14645
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2534e434cd5abbb8f7668d3eab0549db0ef95d6a797a3efa86b712e8e32266a7
|
| 3 |
size 14645
|
scheduler.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 1465
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f35223b4162b3f25fe602e5e4c5a2349c08c0134f11cd20a82d190f37cb0842a
|
| 3 |
size 1465
|
trainer_state.json
CHANGED
|
@@ -2,666 +2,1016 @@
|
|
| 2 |
"best_global_step": null,
|
| 3 |
"best_metric": null,
|
| 4 |
"best_model_checkpoint": null,
|
| 5 |
-
"epoch":
|
| 6 |
"eval_steps": 500,
|
| 7 |
-
"global_step":
|
| 8 |
"is_hyper_param_search": false,
|
| 9 |
"is_local_process_zero": true,
|
| 10 |
"is_world_process_zero": true,
|
| 11 |
"log_history": [
|
| 12 |
{
|
| 13 |
-
"entropy": 2.
|
| 14 |
-
"epoch": 0.
|
| 15 |
-
"grad_norm":
|
| 16 |
"learning_rate": 0.0,
|
| 17 |
-
"loss": 2.
|
| 18 |
-
"mean_token_accuracy": 0.
|
| 19 |
-
"num_tokens":
|
| 20 |
"step": 1
|
| 21 |
},
|
| 22 |
{
|
| 23 |
-
"entropy": 2.
|
| 24 |
-
"epoch": 0.
|
| 25 |
-
"grad_norm":
|
| 26 |
"learning_rate": 2e-06,
|
| 27 |
-
"loss": 2.
|
| 28 |
-
"mean_token_accuracy": 0.
|
| 29 |
-
"num_tokens":
|
| 30 |
"step": 2
|
| 31 |
},
|
| 32 |
{
|
| 33 |
-
"entropy": 2.
|
| 34 |
-
"epoch": 0.
|
| 35 |
-
"grad_norm":
|
| 36 |
"learning_rate": 4e-06,
|
| 37 |
-
"loss":
|
| 38 |
-
"mean_token_accuracy": 0.
|
| 39 |
-
"num_tokens":
|
| 40 |
"step": 3
|
| 41 |
},
|
| 42 |
{
|
| 43 |
-
"entropy": 2.
|
| 44 |
-
"epoch": 0.
|
| 45 |
-
"grad_norm":
|
| 46 |
"learning_rate": 6e-06,
|
| 47 |
-
"loss": 2.
|
| 48 |
-
"mean_token_accuracy": 0.
|
| 49 |
-
"num_tokens":
|
| 50 |
"step": 4
|
| 51 |
},
|
| 52 |
{
|
| 53 |
-
"entropy": 2.
|
| 54 |
-
"epoch": 0.
|
| 55 |
-
"grad_norm":
|
| 56 |
"learning_rate": 8e-06,
|
| 57 |
-
"loss":
|
| 58 |
-
"mean_token_accuracy": 0.
|
| 59 |
-
"num_tokens":
|
| 60 |
"step": 5
|
| 61 |
},
|
| 62 |
{
|
| 63 |
-
"entropy": 2.
|
| 64 |
-
"epoch": 0.
|
| 65 |
-
"grad_norm":
|
| 66 |
"learning_rate": 9.999999999999999e-06,
|
| 67 |
-
"loss": 2.
|
| 68 |
-
"mean_token_accuracy": 0.
|
| 69 |
-
"num_tokens":
|
| 70 |
"step": 6
|
| 71 |
},
|
| 72 |
{
|
| 73 |
-
"entropy": 2.
|
| 74 |
-
"epoch": 0.
|
| 75 |
-
"grad_norm":
|
| 76 |
"learning_rate": 1.2e-05,
|
| 77 |
-
"loss":
|
| 78 |
-
"mean_token_accuracy": 0.
|
| 79 |
-
"num_tokens":
|
| 80 |
"step": 7
|
| 81 |
},
|
| 82 |
{
|
| 83 |
-
"entropy": 2.
|
| 84 |
-
"epoch": 0.
|
| 85 |
-
"grad_norm":
|
| 86 |
"learning_rate": 1.4e-05,
|
| 87 |
-
"loss": 1.
|
| 88 |
-
"mean_token_accuracy": 0.
|
| 89 |
-
"num_tokens":
|
| 90 |
"step": 8
|
| 91 |
},
|
| 92 |
{
|
| 93 |
-
"entropy": 2.
|
| 94 |
-
"epoch": 0.
|
| 95 |
-
"grad_norm":
|
| 96 |
"learning_rate": 1.6e-05,
|
| 97 |
-
"loss": 1.
|
| 98 |
-
"mean_token_accuracy": 0.
|
| 99 |
-
"num_tokens":
|
| 100 |
"step": 9
|
| 101 |
},
|
| 102 |
{
|
| 103 |
-
"entropy": 2.
|
| 104 |
-
"epoch": 0.
|
| 105 |
-
"grad_norm":
|
| 106 |
"learning_rate": 1.8e-05,
|
| 107 |
-
"loss": 1.
|
| 108 |
-
"mean_token_accuracy": 0.
|
| 109 |
-
"num_tokens":
|
| 110 |
"step": 10
|
| 111 |
},
|
| 112 |
{
|
| 113 |
-
"entropy":
|
| 114 |
-
"epoch": 0.
|
| 115 |
-
"grad_norm":
|
| 116 |
"learning_rate": 1.9999999999999998e-05,
|
| 117 |
-
"loss": 1.
|
| 118 |
-
"mean_token_accuracy": 0.
|
| 119 |
-
"num_tokens":
|
| 120 |
"step": 11
|
| 121 |
},
|
| 122 |
{
|
| 123 |
-
"entropy": 2.
|
| 124 |
-
"epoch": 0.
|
| 125 |
-
"grad_norm":
|
| 126 |
"learning_rate": 2.2e-05,
|
| 127 |
-
"loss":
|
| 128 |
-
"mean_token_accuracy": 0.
|
| 129 |
-
"num_tokens":
|
| 130 |
"step": 12
|
| 131 |
},
|
| 132 |
{
|
| 133 |
-
"entropy": 2.
|
| 134 |
-
"epoch": 0.
|
| 135 |
-
"grad_norm":
|
| 136 |
"learning_rate": 2.4e-05,
|
| 137 |
-
"loss": 1.
|
| 138 |
-
"mean_token_accuracy": 0.
|
| 139 |
-
"num_tokens":
|
| 140 |
"step": 13
|
| 141 |
},
|
| 142 |
{
|
| 143 |
-
"entropy": 2.
|
| 144 |
-
"epoch": 0.
|
| 145 |
-
"grad_norm":
|
| 146 |
"learning_rate": 2.6000000000000002e-05,
|
| 147 |
-
"loss": 1.
|
| 148 |
-
"mean_token_accuracy": 0.
|
| 149 |
-
"num_tokens":
|
| 150 |
"step": 14
|
| 151 |
},
|
| 152 |
{
|
| 153 |
-
"entropy": 2.
|
| 154 |
-
"epoch": 0.
|
| 155 |
-
"grad_norm":
|
| 156 |
"learning_rate": 2.8e-05,
|
| 157 |
-
"loss": 1.
|
| 158 |
-
"mean_token_accuracy": 0.
|
| 159 |
-
"num_tokens":
|
| 160 |
"step": 15
|
| 161 |
},
|
| 162 |
{
|
| 163 |
-
"entropy": 2.
|
| 164 |
-
"epoch": 0.
|
| 165 |
-
"grad_norm":
|
| 166 |
"learning_rate": 3e-05,
|
| 167 |
-
"loss": 1.
|
| 168 |
-
"mean_token_accuracy": 0.
|
| 169 |
-
"num_tokens":
|
| 170 |
"step": 16
|
| 171 |
},
|
| 172 |
{
|
| 173 |
-
"entropy": 2.
|
| 174 |
-
"epoch": 0.
|
| 175 |
-
"grad_norm":
|
| 176 |
-
"learning_rate": 2.
|
| 177 |
-
"loss": 1.
|
| 178 |
-
"mean_token_accuracy": 0.
|
| 179 |
-
"num_tokens":
|
| 180 |
"step": 17
|
| 181 |
},
|
| 182 |
{
|
| 183 |
-
"entropy":
|
| 184 |
-
"epoch": 0.
|
| 185 |
-
"grad_norm":
|
| 186 |
-
"learning_rate": 2.
|
| 187 |
-
"loss": 1.
|
| 188 |
-
"mean_token_accuracy": 0.
|
| 189 |
-
"num_tokens":
|
| 190 |
"step": 18
|
| 191 |
},
|
| 192 |
{
|
| 193 |
-
"entropy":
|
| 194 |
-
"epoch": 0.
|
| 195 |
-
"grad_norm":
|
| 196 |
-
"learning_rate": 2.
|
| 197 |
-
"loss": 1.
|
| 198 |
-
"mean_token_accuracy": 0.
|
| 199 |
-
"num_tokens":
|
| 200 |
"step": 19
|
| 201 |
},
|
| 202 |
{
|
| 203 |
-
"entropy": 1.
|
| 204 |
-
"epoch": 0.
|
| 205 |
-
"grad_norm":
|
| 206 |
-
"learning_rate": 2.
|
| 207 |
-
"loss": 1.
|
| 208 |
-
"mean_token_accuracy": 0.
|
| 209 |
-
"num_tokens":
|
| 210 |
"step": 20
|
| 211 |
},
|
| 212 |
{
|
| 213 |
-
"entropy":
|
| 214 |
-
"epoch": 0.
|
| 215 |
-
"grad_norm":
|
| 216 |
-
"learning_rate": 2.
|
| 217 |
-
"loss": 1.
|
| 218 |
-
"mean_token_accuracy": 0.
|
| 219 |
-
"num_tokens":
|
| 220 |
"step": 21
|
| 221 |
},
|
| 222 |
{
|
| 223 |
-
"entropy":
|
| 224 |
-
"epoch": 0.
|
| 225 |
-
"grad_norm":
|
| 226 |
-
"learning_rate": 2.
|
| 227 |
-
"loss": 1.
|
| 228 |
-
"mean_token_accuracy": 0.
|
| 229 |
-
"num_tokens":
|
| 230 |
"step": 22
|
| 231 |
},
|
| 232 |
{
|
| 233 |
-
"entropy": 1.
|
| 234 |
-
"epoch": 0.
|
| 235 |
-
"grad_norm":
|
| 236 |
-
"learning_rate": 2.
|
| 237 |
-
"loss": 1.
|
| 238 |
-
"mean_token_accuracy": 0.
|
| 239 |
-
"num_tokens":
|
| 240 |
"step": 23
|
| 241 |
},
|
| 242 |
{
|
| 243 |
-
"entropy": 2.
|
| 244 |
-
"epoch": 0.
|
| 245 |
-
"grad_norm":
|
| 246 |
-
"learning_rate": 2.
|
| 247 |
-
"loss": 1.
|
| 248 |
-
"mean_token_accuracy": 0.
|
| 249 |
-
"num_tokens":
|
| 250 |
"step": 24
|
| 251 |
},
|
| 252 |
{
|
| 253 |
-
"entropy":
|
| 254 |
-
"epoch": 0.
|
| 255 |
-
"grad_norm":
|
| 256 |
-
"learning_rate": 2.
|
| 257 |
-
"loss": 1.
|
| 258 |
-
"mean_token_accuracy": 0.
|
| 259 |
-
"num_tokens":
|
| 260 |
"step": 25
|
| 261 |
},
|
| 262 |
{
|
| 263 |
-
"entropy":
|
| 264 |
-
"epoch": 0.
|
| 265 |
-
"grad_norm":
|
| 266 |
-
"learning_rate": 2.
|
| 267 |
-
"loss": 1.
|
| 268 |
-
"mean_token_accuracy": 0.
|
| 269 |
-
"num_tokens":
|
| 270 |
"step": 26
|
| 271 |
},
|
| 272 |
{
|
| 273 |
-
"entropy":
|
| 274 |
-
"epoch": 0.
|
| 275 |
-
"grad_norm":
|
| 276 |
-
"learning_rate": 2.
|
| 277 |
-
"loss": 1.
|
| 278 |
-
"mean_token_accuracy": 0.
|
| 279 |
-
"num_tokens":
|
| 280 |
"step": 27
|
| 281 |
},
|
| 282 |
{
|
| 283 |
-
"entropy":
|
| 284 |
-
"epoch": 0.
|
| 285 |
-
"grad_norm":
|
| 286 |
-
"learning_rate": 2.
|
| 287 |
-
"loss": 1.
|
| 288 |
-
"mean_token_accuracy": 0.
|
| 289 |
-
"num_tokens":
|
| 290 |
"step": 28
|
| 291 |
},
|
| 292 |
{
|
| 293 |
-
"entropy": 2.
|
| 294 |
-
"epoch": 0.
|
| 295 |
-
"grad_norm":
|
| 296 |
-
"learning_rate": 2.
|
| 297 |
-
"loss": 1.
|
| 298 |
-
"mean_token_accuracy": 0.
|
| 299 |
-
"num_tokens":
|
| 300 |
"step": 29
|
| 301 |
},
|
| 302 |
{
|
| 303 |
-
"entropy": 2.
|
| 304 |
-
"epoch": 0.
|
| 305 |
-
"grad_norm":
|
| 306 |
-
"learning_rate": 2.
|
| 307 |
-
"loss": 1.
|
| 308 |
-
"mean_token_accuracy": 0.
|
| 309 |
-
"num_tokens":
|
| 310 |
"step": 30
|
| 311 |
},
|
| 312 |
{
|
| 313 |
-
"entropy":
|
| 314 |
-
"epoch": 0.
|
| 315 |
-
"grad_norm":
|
| 316 |
-
"learning_rate": 2.
|
| 317 |
-
"loss": 1.
|
| 318 |
-
"mean_token_accuracy": 0.
|
| 319 |
-
"num_tokens":
|
| 320 |
"step": 31
|
| 321 |
},
|
| 322 |
{
|
| 323 |
-
"entropy":
|
| 324 |
-
"epoch": 0.
|
| 325 |
-
"grad_norm":
|
| 326 |
-
"learning_rate": 2.
|
| 327 |
-
"loss": 1.
|
| 328 |
-
"mean_token_accuracy": 0.
|
| 329 |
-
"num_tokens":
|
| 330 |
"step": 32
|
| 331 |
},
|
| 332 |
{
|
| 333 |
-
"entropy": 1.
|
| 334 |
-
"epoch":
|
| 335 |
-
"grad_norm":
|
| 336 |
-
"learning_rate": 2.
|
| 337 |
-
"loss": 1.
|
| 338 |
-
"mean_token_accuracy": 0.
|
| 339 |
-
"num_tokens":
|
| 340 |
"step": 33
|
| 341 |
},
|
| 342 |
{
|
| 343 |
-
"entropy": 1.
|
| 344 |
-
"epoch":
|
| 345 |
-
"grad_norm":
|
| 346 |
-
"learning_rate": 2.
|
| 347 |
-
"loss":
|
| 348 |
-
"mean_token_accuracy": 0.
|
| 349 |
-
"num_tokens":
|
| 350 |
"step": 34
|
| 351 |
},
|
| 352 |
{
|
| 353 |
-
"entropy": 1.
|
| 354 |
-
"epoch": 1.
|
| 355 |
-
"grad_norm":
|
| 356 |
-
"learning_rate": 2.
|
| 357 |
-
"loss": 1.
|
| 358 |
-
"mean_token_accuracy": 0.
|
| 359 |
-
"num_tokens":
|
| 360 |
"step": 35
|
| 361 |
},
|
| 362 |
{
|
| 363 |
-
"entropy": 1.
|
| 364 |
-
"epoch": 1.
|
| 365 |
-
"grad_norm":
|
| 366 |
-
"learning_rate": 2.
|
| 367 |
-
"loss":
|
| 368 |
-
"mean_token_accuracy": 0.
|
| 369 |
-
"num_tokens":
|
| 370 |
"step": 36
|
| 371 |
},
|
| 372 |
{
|
| 373 |
-
"entropy": 1.
|
| 374 |
-
"epoch": 1.
|
| 375 |
-
"grad_norm":
|
| 376 |
-
"learning_rate": 2.
|
| 377 |
-
"loss":
|
| 378 |
-
"mean_token_accuracy": 0.
|
| 379 |
-
"num_tokens":
|
| 380 |
"step": 37
|
| 381 |
},
|
| 382 |
{
|
| 383 |
-
"entropy": 1.
|
| 384 |
-
"epoch": 1.
|
| 385 |
-
"grad_norm":
|
| 386 |
-
"learning_rate": 2.
|
| 387 |
-
"loss":
|
| 388 |
-
"mean_token_accuracy": 0.
|
| 389 |
-
"num_tokens":
|
| 390 |
"step": 38
|
| 391 |
},
|
| 392 |
{
|
| 393 |
-
"entropy": 1.
|
| 394 |
-
"epoch": 1.
|
| 395 |
-
"grad_norm":
|
| 396 |
-
"learning_rate": 2.
|
| 397 |
-
"loss":
|
| 398 |
-
"mean_token_accuracy": 0.
|
| 399 |
-
"num_tokens":
|
| 400 |
"step": 39
|
| 401 |
},
|
| 402 |
{
|
| 403 |
-
"entropy": 1.
|
| 404 |
-
"epoch": 1.
|
| 405 |
-
"grad_norm":
|
| 406 |
-
"learning_rate": 2.
|
| 407 |
-
"loss":
|
| 408 |
-
"mean_token_accuracy": 0.
|
| 409 |
-
"num_tokens":
|
| 410 |
"step": 40
|
| 411 |
},
|
| 412 |
{
|
| 413 |
-
"entropy": 1.
|
| 414 |
-
"epoch": 1.
|
| 415 |
-
"grad_norm":
|
| 416 |
-
"learning_rate": 2.
|
| 417 |
-
"loss":
|
| 418 |
-
"mean_token_accuracy": 0.
|
| 419 |
-
"num_tokens":
|
| 420 |
"step": 41
|
| 421 |
},
|
| 422 |
{
|
| 423 |
-
"entropy": 1.
|
| 424 |
-
"epoch": 1.
|
| 425 |
-
"grad_norm":
|
| 426 |
-
"learning_rate": 2.
|
| 427 |
-
"loss":
|
| 428 |
-
"mean_token_accuracy": 0.
|
| 429 |
-
"num_tokens":
|
| 430 |
"step": 42
|
| 431 |
},
|
| 432 |
{
|
| 433 |
-
"entropy": 1.
|
| 434 |
-
"epoch": 1.
|
| 435 |
-
"grad_norm":
|
| 436 |
-
"learning_rate": 2.
|
| 437 |
-
"loss":
|
| 438 |
-
"mean_token_accuracy": 0.
|
| 439 |
-
"num_tokens":
|
| 440 |
"step": 43
|
| 441 |
},
|
| 442 |
{
|
| 443 |
-
"entropy": 1.
|
| 444 |
-
"epoch": 1.
|
| 445 |
-
"grad_norm":
|
| 446 |
-
"learning_rate": 2.
|
| 447 |
-
"loss":
|
| 448 |
-
"mean_token_accuracy": 0.
|
| 449 |
-
"num_tokens":
|
| 450 |
"step": 44
|
| 451 |
},
|
| 452 |
{
|
| 453 |
-
"entropy": 1.
|
| 454 |
-
"epoch": 1.
|
| 455 |
-
"grad_norm":
|
| 456 |
-
"learning_rate": 2.
|
| 457 |
-
"loss":
|
| 458 |
-
"mean_token_accuracy": 0.
|
| 459 |
-
"num_tokens":
|
| 460 |
"step": 45
|
| 461 |
},
|
| 462 |
{
|
| 463 |
-
"entropy": 1.
|
| 464 |
-
"epoch": 1.
|
| 465 |
-
"grad_norm":
|
| 466 |
-
"learning_rate": 2.
|
| 467 |
-
"loss":
|
| 468 |
-
"mean_token_accuracy": 0.
|
| 469 |
-
"num_tokens":
|
| 470 |
"step": 46
|
| 471 |
},
|
| 472 |
{
|
| 473 |
-
"entropy": 1.
|
| 474 |
-
"epoch": 1.
|
| 475 |
-
"grad_norm":
|
| 476 |
-
"learning_rate": 2.
|
| 477 |
-
"loss":
|
| 478 |
-
"mean_token_accuracy": 0.
|
| 479 |
-
"num_tokens":
|
| 480 |
"step": 47
|
| 481 |
},
|
| 482 |
{
|
| 483 |
-
"entropy": 1.
|
| 484 |
-
"epoch": 1.
|
| 485 |
-
"grad_norm":
|
| 486 |
-
"learning_rate": 2.
|
| 487 |
-
"loss":
|
| 488 |
-
"mean_token_accuracy": 0.
|
| 489 |
-
"num_tokens":
|
| 490 |
"step": 48
|
| 491 |
},
|
| 492 |
{
|
| 493 |
-
"entropy": 1.
|
| 494 |
-
"epoch": 1.
|
| 495 |
-
"grad_norm":
|
| 496 |
-
"learning_rate":
|
| 497 |
-
"loss":
|
| 498 |
-
"mean_token_accuracy": 0.
|
| 499 |
-
"num_tokens":
|
| 500 |
"step": 49
|
| 501 |
},
|
| 502 |
{
|
| 503 |
-
"entropy": 1.
|
| 504 |
-
"epoch": 1.
|
| 505 |
-
"grad_norm":
|
| 506 |
-
"learning_rate":
|
| 507 |
-
"loss":
|
| 508 |
-
"mean_token_accuracy": 0.
|
| 509 |
-
"num_tokens":
|
| 510 |
"step": 50
|
| 511 |
},
|
| 512 |
{
|
| 513 |
-
"entropy": 1.
|
| 514 |
-
"epoch": 1.
|
| 515 |
-
"grad_norm":
|
| 516 |
-
"learning_rate":
|
| 517 |
-
"loss":
|
| 518 |
-
"mean_token_accuracy": 0.
|
| 519 |
-
"num_tokens":
|
| 520 |
"step": 51
|
| 521 |
},
|
| 522 |
{
|
| 523 |
-
"entropy": 1.
|
| 524 |
-
"epoch": 1.
|
| 525 |
-
"grad_norm":
|
| 526 |
-
"learning_rate": 1.
|
| 527 |
-
"loss":
|
| 528 |
-
"mean_token_accuracy": 0.
|
| 529 |
-
"num_tokens":
|
| 530 |
"step": 52
|
| 531 |
},
|
| 532 |
{
|
| 533 |
-
"entropy": 1.
|
| 534 |
-
"epoch": 1.
|
| 535 |
-
"grad_norm":
|
| 536 |
-
"learning_rate": 1.
|
| 537 |
-
"loss":
|
| 538 |
-
"mean_token_accuracy": 0.
|
| 539 |
-
"num_tokens":
|
| 540 |
"step": 53
|
| 541 |
},
|
| 542 |
{
|
| 543 |
-
"entropy": 1.
|
| 544 |
-
"epoch": 1.
|
| 545 |
-
"grad_norm":
|
| 546 |
-
"learning_rate": 1.
|
| 547 |
-
"loss":
|
| 548 |
-
"mean_token_accuracy": 0.
|
| 549 |
-
"num_tokens":
|
| 550 |
"step": 54
|
| 551 |
},
|
| 552 |
{
|
| 553 |
-
"entropy": 1.
|
| 554 |
-
"epoch": 1.
|
| 555 |
-
"grad_norm":
|
| 556 |
-
"learning_rate": 1.
|
| 557 |
-
"loss":
|
| 558 |
-
"mean_token_accuracy": 0.
|
| 559 |
-
"num_tokens":
|
| 560 |
"step": 55
|
| 561 |
},
|
| 562 |
{
|
| 563 |
-
"entropy": 1.
|
| 564 |
-
"epoch": 1.
|
| 565 |
-
"grad_norm":
|
| 566 |
-
"learning_rate": 1.
|
| 567 |
-
"loss":
|
| 568 |
-
"mean_token_accuracy": 0.
|
| 569 |
-
"num_tokens":
|
| 570 |
"step": 56
|
| 571 |
},
|
| 572 |
{
|
| 573 |
-
"entropy": 1.
|
| 574 |
-
"epoch": 1.
|
| 575 |
-
"grad_norm":
|
| 576 |
-
"learning_rate": 1.
|
| 577 |
-
"loss":
|
| 578 |
-
"mean_token_accuracy": 0.
|
| 579 |
-
"num_tokens":
|
| 580 |
"step": 57
|
| 581 |
},
|
| 582 |
{
|
| 583 |
-
"entropy": 1.
|
| 584 |
-
"epoch": 1.
|
| 585 |
-
"grad_norm":
|
| 586 |
-
"learning_rate": 1.
|
| 587 |
-
"loss":
|
| 588 |
-
"mean_token_accuracy": 0.
|
| 589 |
-
"num_tokens":
|
| 590 |
"step": 58
|
| 591 |
},
|
| 592 |
{
|
| 593 |
-
"entropy": 1.
|
| 594 |
-
"epoch": 1.
|
| 595 |
-
"grad_norm":
|
| 596 |
-
"learning_rate": 1.
|
| 597 |
-
"loss":
|
| 598 |
-
"mean_token_accuracy": 0.
|
| 599 |
-
"num_tokens":
|
| 600 |
"step": 59
|
| 601 |
},
|
| 602 |
{
|
| 603 |
-
"entropy": 1.
|
| 604 |
-
"epoch": 1.
|
| 605 |
-
"grad_norm":
|
| 606 |
-
"learning_rate": 1.
|
| 607 |
-
"loss":
|
| 608 |
-
"mean_token_accuracy": 0.
|
| 609 |
-
"num_tokens":
|
| 610 |
"step": 60
|
| 611 |
},
|
| 612 |
{
|
| 613 |
-
"entropy": 1.
|
| 614 |
-
"epoch": 1.
|
| 615 |
-
"grad_norm":
|
| 616 |
-
"learning_rate": 1.
|
| 617 |
-
"loss":
|
| 618 |
-
"mean_token_accuracy": 0.
|
| 619 |
-
"num_tokens":
|
| 620 |
"step": 61
|
| 621 |
},
|
| 622 |
{
|
| 623 |
-
"entropy": 1.
|
| 624 |
-
"epoch": 1.
|
| 625 |
-
"grad_norm":
|
| 626 |
-
"learning_rate": 1.
|
| 627 |
-
"loss": 1.
|
| 628 |
-
"mean_token_accuracy": 0.
|
| 629 |
-
"num_tokens":
|
| 630 |
"step": 62
|
| 631 |
},
|
| 632 |
{
|
| 633 |
-
"entropy": 1.
|
| 634 |
-
"epoch": 1.
|
| 635 |
-
"grad_norm":
|
| 636 |
-
"learning_rate": 1.
|
| 637 |
-
"loss":
|
| 638 |
-
"mean_token_accuracy": 0.
|
| 639 |
-
"num_tokens":
|
| 640 |
"step": 63
|
| 641 |
},
|
| 642 |
{
|
| 643 |
-
"entropy": 1.
|
| 644 |
-
"epoch": 1.
|
| 645 |
-
"grad_norm":
|
| 646 |
-
"learning_rate": 1.
|
| 647 |
-
"loss":
|
| 648 |
-
"mean_token_accuracy": 0.
|
| 649 |
-
"num_tokens":
|
| 650 |
"step": 64
|
| 651 |
},
|
| 652 |
{
|
| 653 |
-
"entropy": 1.
|
| 654 |
-
"epoch": 1.
|
| 655 |
-
"grad_norm":
|
| 656 |
-
"learning_rate": 1.
|
| 657 |
-
"loss":
|
| 658 |
-
"mean_token_accuracy": 0.
|
| 659 |
-
"num_tokens":
|
| 660 |
"step": 65
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 661 |
}
|
| 662 |
],
|
| 663 |
"logging_steps": 1,
|
| 664 |
-
"max_steps":
|
| 665 |
"num_input_tokens_seen": 0,
|
| 666 |
"num_train_epochs": 3,
|
| 667 |
"save_steps": 5,
|
|
@@ -677,7 +1027,7 @@
|
|
| 677 |
"attributes": {}
|
| 678 |
}
|
| 679 |
},
|
| 680 |
-
"total_flos":
|
| 681 |
"train_batch_size": 2,
|
| 682 |
"trial_name": null,
|
| 683 |
"trial_params": null
|
|
|
|
| 2 |
"best_global_step": null,
|
| 3 |
"best_metric": null,
|
| 4 |
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 2.875912408759124,
|
| 6 |
"eval_steps": 500,
|
| 7 |
+
"global_step": 100,
|
| 8 |
"is_hyper_param_search": false,
|
| 9 |
"is_local_process_zero": true,
|
| 10 |
"is_world_process_zero": true,
|
| 11 |
"log_history": [
|
| 12 |
{
|
| 13 |
+
"entropy": 2.134004071354866,
|
| 14 |
+
"epoch": 0.029197080291970802,
|
| 15 |
+
"grad_norm": 19.125,
|
| 16 |
"learning_rate": 0.0,
|
| 17 |
+
"loss": 2.5766,
|
| 18 |
+
"mean_token_accuracy": 0.42293117567896843,
|
| 19 |
+
"num_tokens": 1699.0,
|
| 20 |
"step": 1
|
| 21 |
},
|
| 22 |
{
|
| 23 |
+
"entropy": 2.1332614570856094,
|
| 24 |
+
"epoch": 0.058394160583941604,
|
| 25 |
+
"grad_norm": 15.375,
|
| 26 |
"learning_rate": 2e-06,
|
| 27 |
+
"loss": 2.3501,
|
| 28 |
+
"mean_token_accuracy": 0.43237315863370895,
|
| 29 |
+
"num_tokens": 3890.0,
|
| 30 |
"step": 2
|
| 31 |
},
|
| 32 |
{
|
| 33 |
+
"entropy": 2.3441822230815887,
|
| 34 |
+
"epoch": 0.08759124087591241,
|
| 35 |
+
"grad_norm": 22.875,
|
| 36 |
"learning_rate": 4e-06,
|
| 37 |
+
"loss": 3.134,
|
| 38 |
+
"mean_token_accuracy": 0.3771548382937908,
|
| 39 |
+
"num_tokens": 5114.0,
|
| 40 |
"step": 3
|
| 41 |
},
|
| 42 |
{
|
| 43 |
+
"entropy": 2.169568419456482,
|
| 44 |
+
"epoch": 0.11678832116788321,
|
| 45 |
+
"grad_norm": 15.0625,
|
| 46 |
"learning_rate": 6e-06,
|
| 47 |
+
"loss": 2.2609,
|
| 48 |
+
"mean_token_accuracy": 0.4582384452223778,
|
| 49 |
+
"num_tokens": 6825.0,
|
| 50 |
"step": 4
|
| 51 |
},
|
| 52 |
{
|
| 53 |
+
"entropy": 2.3993491530418396,
|
| 54 |
+
"epoch": 0.145985401459854,
|
| 55 |
+
"grad_norm": 11.5625,
|
| 56 |
"learning_rate": 8e-06,
|
| 57 |
+
"loss": 2.2625,
|
| 58 |
+
"mean_token_accuracy": 0.44751258939504623,
|
| 59 |
+
"num_tokens": 8794.0,
|
| 60 |
"step": 5
|
| 61 |
},
|
| 62 |
{
|
| 63 |
+
"entropy": 2.38617005944252,
|
| 64 |
+
"epoch": 0.17518248175182483,
|
| 65 |
+
"grad_norm": 11.0,
|
| 66 |
"learning_rate": 9.999999999999999e-06,
|
| 67 |
+
"loss": 2.2774,
|
| 68 |
+
"mean_token_accuracy": 0.4528072811663151,
|
| 69 |
+
"num_tokens": 10473.0,
|
| 70 |
"step": 6
|
| 71 |
},
|
| 72 |
{
|
| 73 |
+
"entropy": 2.17643141746521,
|
| 74 |
+
"epoch": 0.20437956204379562,
|
| 75 |
+
"grad_norm": 7.21875,
|
| 76 |
"learning_rate": 1.2e-05,
|
| 77 |
+
"loss": 1.911,
|
| 78 |
+
"mean_token_accuracy": 0.5113650299608707,
|
| 79 |
+
"num_tokens": 12709.0,
|
| 80 |
"step": 7
|
| 81 |
},
|
| 82 |
{
|
| 83 |
+
"entropy": 2.2615339010953903,
|
| 84 |
+
"epoch": 0.23357664233576642,
|
| 85 |
+
"grad_norm": 5.9375,
|
| 86 |
"learning_rate": 1.4e-05,
|
| 87 |
+
"loss": 1.8747,
|
| 88 |
+
"mean_token_accuracy": 0.5006480813026428,
|
| 89 |
+
"num_tokens": 15657.0,
|
| 90 |
"step": 8
|
| 91 |
},
|
| 92 |
{
|
| 93 |
+
"entropy": 2.416978284716606,
|
| 94 |
+
"epoch": 0.26277372262773724,
|
| 95 |
+
"grad_norm": 8.4375,
|
| 96 |
"learning_rate": 1.6e-05,
|
| 97 |
+
"loss": 1.9924,
|
| 98 |
+
"mean_token_accuracy": 0.4901970997452736,
|
| 99 |
+
"num_tokens": 17681.0,
|
| 100 |
"step": 9
|
| 101 |
},
|
| 102 |
{
|
| 103 |
+
"entropy": 2.3273025155067444,
|
| 104 |
+
"epoch": 0.291970802919708,
|
| 105 |
+
"grad_norm": 6.09375,
|
| 106 |
"learning_rate": 1.8e-05,
|
| 107 |
+
"loss": 1.8238,
|
| 108 |
+
"mean_token_accuracy": 0.4891773872077465,
|
| 109 |
+
"num_tokens": 20159.0,
|
| 110 |
"step": 10
|
| 111 |
},
|
| 112 |
{
|
| 113 |
+
"entropy": 2.295111373066902,
|
| 114 |
+
"epoch": 0.32116788321167883,
|
| 115 |
+
"grad_norm": 5.9375,
|
| 116 |
"learning_rate": 1.9999999999999998e-05,
|
| 117 |
+
"loss": 1.8838,
|
| 118 |
+
"mean_token_accuracy": 0.528899259865284,
|
| 119 |
+
"num_tokens": 22380.0,
|
| 120 |
"step": 11
|
| 121 |
},
|
| 122 |
{
|
| 123 |
+
"entropy": 2.4463636726140976,
|
| 124 |
+
"epoch": 0.35036496350364965,
|
| 125 |
+
"grad_norm": 7.28125,
|
| 126 |
"learning_rate": 2.2e-05,
|
| 127 |
+
"loss": 2.0672,
|
| 128 |
+
"mean_token_accuracy": 0.4942050985991955,
|
| 129 |
+
"num_tokens": 23949.0,
|
| 130 |
"step": 12
|
| 131 |
},
|
| 132 |
{
|
| 133 |
+
"entropy": 2.2411956042051315,
|
| 134 |
+
"epoch": 0.3795620437956204,
|
| 135 |
+
"grad_norm": 6.625,
|
| 136 |
"learning_rate": 2.4e-05,
|
| 137 |
+
"loss": 1.7258,
|
| 138 |
+
"mean_token_accuracy": 0.5641119256615639,
|
| 139 |
+
"num_tokens": 25626.0,
|
| 140 |
"step": 13
|
| 141 |
},
|
| 142 |
{
|
| 143 |
+
"entropy": 2.1571693122386932,
|
| 144 |
+
"epoch": 0.40875912408759124,
|
| 145 |
+
"grad_norm": 6.3125,
|
| 146 |
"learning_rate": 2.6000000000000002e-05,
|
| 147 |
+
"loss": 1.7421,
|
| 148 |
+
"mean_token_accuracy": 0.5413074977695942,
|
| 149 |
+
"num_tokens": 27703.0,
|
| 150 |
"step": 14
|
| 151 |
},
|
| 152 |
{
|
| 153 |
+
"entropy": 2.0649050027132034,
|
| 154 |
+
"epoch": 0.43795620437956206,
|
| 155 |
+
"grad_norm": 5.65625,
|
| 156 |
"learning_rate": 2.8e-05,
|
| 157 |
+
"loss": 1.7653,
|
| 158 |
+
"mean_token_accuracy": 0.5364297069609165,
|
| 159 |
+
"num_tokens": 29910.0,
|
| 160 |
"step": 15
|
| 161 |
},
|
| 162 |
{
|
| 163 |
+
"entropy": 2.0259645730257034,
|
| 164 |
+
"epoch": 0.46715328467153283,
|
| 165 |
+
"grad_norm": 5.5,
|
| 166 |
"learning_rate": 3e-05,
|
| 167 |
+
"loss": 1.586,
|
| 168 |
+
"mean_token_accuracy": 0.5716114267706871,
|
| 169 |
+
"num_tokens": 32243.0,
|
| 170 |
"step": 16
|
| 171 |
},
|
| 172 |
{
|
| 173 |
+
"entropy": 2.2259650826454163,
|
| 174 |
+
"epoch": 0.49635036496350365,
|
| 175 |
+
"grad_norm": 6.46875,
|
| 176 |
+
"learning_rate": 2.9990862405286438e-05,
|
| 177 |
+
"loss": 1.8815,
|
| 178 |
+
"mean_token_accuracy": 0.5339390859007835,
|
| 179 |
+
"num_tokens": 33704.0,
|
| 180 |
"step": 17
|
| 181 |
},
|
| 182 |
{
|
| 183 |
+
"entropy": 2.2045857161283493,
|
| 184 |
+
"epoch": 0.5255474452554745,
|
| 185 |
+
"grad_norm": 7.0,
|
| 186 |
+
"learning_rate": 2.9963460753897364e-05,
|
| 187 |
+
"loss": 1.8033,
|
| 188 |
+
"mean_token_accuracy": 0.5426613725721836,
|
| 189 |
+
"num_tokens": 35222.0,
|
| 190 |
"step": 18
|
| 191 |
},
|
| 192 |
{
|
| 193 |
+
"entropy": 2.0502880662679672,
|
| 194 |
+
"epoch": 0.5547445255474452,
|
| 195 |
+
"grad_norm": 5.625,
|
| 196 |
+
"learning_rate": 2.99178284305241e-05,
|
| 197 |
+
"loss": 1.6822,
|
| 198 |
+
"mean_token_accuracy": 0.5445077642798424,
|
| 199 |
+
"num_tokens": 37281.0,
|
| 200 |
"step": 19
|
| 201 |
},
|
| 202 |
{
|
| 203 |
+
"entropy": 1.9275199472904205,
|
| 204 |
+
"epoch": 0.583941605839416,
|
| 205 |
+
"grad_norm": 5.625,
|
| 206 |
+
"learning_rate": 2.9854021031123555e-05,
|
| 207 |
+
"loss": 1.5569,
|
| 208 |
+
"mean_token_accuracy": 0.5689515694975853,
|
| 209 |
+
"num_tokens": 39208.0,
|
| 210 |
"step": 20
|
| 211 |
},
|
| 212 |
{
|
| 213 |
+
"entropy": 2.1408673971891403,
|
| 214 |
+
"epoch": 0.6131386861313869,
|
| 215 |
+
"grad_norm": 6.5,
|
| 216 |
+
"learning_rate": 2.977211629518312e-05,
|
| 217 |
+
"loss": 1.8479,
|
| 218 |
+
"mean_token_accuracy": 0.5382610447704792,
|
| 219 |
+
"num_tokens": 40754.0,
|
| 220 |
"step": 21
|
| 221 |
},
|
| 222 |
{
|
| 223 |
+
"entropy": 2.138097256422043,
|
| 224 |
+
"epoch": 0.6423357664233577,
|
| 225 |
+
"grad_norm": 6.03125,
|
| 226 |
+
"learning_rate": 2.9672214011007087e-05,
|
| 227 |
+
"loss": 1.7691,
|
| 228 |
+
"mean_token_accuracy": 0.5337589606642723,
|
| 229 |
+
"num_tokens": 42447.0,
|
| 230 |
"step": 22
|
| 231 |
},
|
| 232 |
{
|
| 233 |
+
"entropy": 1.9584687054157257,
|
| 234 |
+
"epoch": 0.6715328467153284,
|
| 235 |
+
"grad_norm": 4.59375,
|
| 236 |
+
"learning_rate": 2.9554435894139945e-05,
|
| 237 |
+
"loss": 1.502,
|
| 238 |
+
"mean_token_accuracy": 0.5679651834070683,
|
| 239 |
+
"num_tokens": 44963.0,
|
| 240 |
"step": 23
|
| 241 |
},
|
| 242 |
{
|
| 243 |
+
"entropy": 2.0382106602191925,
|
| 244 |
+
"epoch": 0.7007299270072993,
|
| 245 |
+
"grad_norm": 5.03125,
|
| 246 |
+
"learning_rate": 2.9418925439074784e-05,
|
| 247 |
+
"loss": 1.6539,
|
| 248 |
+
"mean_token_accuracy": 0.5411265380680561,
|
| 249 |
+
"num_tokens": 47138.0,
|
| 250 |
"step": 24
|
| 251 |
},
|
| 252 |
{
|
| 253 |
+
"entropy": 2.0515516996383667,
|
| 254 |
+
"epoch": 0.7299270072992701,
|
| 255 |
+
"grad_norm": 5.4375,
|
| 256 |
+
"learning_rate": 2.9265847744427305e-05,
|
| 257 |
+
"loss": 1.7007,
|
| 258 |
+
"mean_token_accuracy": 0.5707135051488876,
|
| 259 |
+
"num_tokens": 49154.0,
|
| 260 |
"step": 25
|
| 261 |
},
|
| 262 |
{
|
| 263 |
+
"entropy": 1.96835595369339,
|
| 264 |
+
"epoch": 0.7591240875912408,
|
| 265 |
+
"grad_norm": 4.875,
|
| 266 |
+
"learning_rate": 2.9095389311788626e-05,
|
| 267 |
+
"loss": 1.5182,
|
| 268 |
+
"mean_token_accuracy": 0.5940572991967201,
|
| 269 |
+
"num_tokens": 51009.0,
|
| 270 |
"step": 26
|
| 271 |
},
|
| 272 |
{
|
| 273 |
+
"entropy": 1.9829230606555939,
|
| 274 |
+
"epoch": 0.7883211678832117,
|
| 275 |
+
"grad_norm": 4.78125,
|
| 276 |
+
"learning_rate": 2.890775781850181e-05,
|
| 277 |
+
"loss": 1.5441,
|
| 278 |
+
"mean_token_accuracy": 0.5696061700582504,
|
| 279 |
+
"num_tokens": 52866.0,
|
| 280 |
"step": 27
|
| 281 |
},
|
| 282 |
{
|
| 283 |
+
"entropy": 1.9901328533887863,
|
| 284 |
+
"epoch": 0.8175182481751825,
|
| 285 |
+
"grad_norm": 4.53125,
|
| 286 |
+
"learning_rate": 2.8703181864639013e-05,
|
| 287 |
+
"loss": 1.5227,
|
| 288 |
+
"mean_token_accuracy": 0.5771616920828819,
|
| 289 |
+
"num_tokens": 55235.0,
|
| 290 |
"step": 28
|
| 291 |
},
|
| 292 |
{
|
| 293 |
+
"entropy": 2.176472947001457,
|
| 294 |
+
"epoch": 0.8467153284671532,
|
| 295 |
+
"grad_norm": 7.34375,
|
| 296 |
+
"learning_rate": 2.8481910694487507e-05,
|
| 297 |
+
"loss": 1.784,
|
| 298 |
+
"mean_token_accuracy": 0.5394799076020718,
|
| 299 |
+
"num_tokens": 56468.0,
|
| 300 |
"step": 29
|
| 301 |
},
|
| 302 |
{
|
| 303 |
+
"entropy": 2.0398730635643005,
|
| 304 |
+
"epoch": 0.8759124087591241,
|
| 305 |
+
"grad_norm": 5.5,
|
| 306 |
+
"learning_rate": 2.8244213892883907e-05,
|
| 307 |
+
"loss": 1.584,
|
| 308 |
+
"mean_token_accuracy": 0.564793273806572,
|
| 309 |
+
"num_tokens": 58219.0,
|
| 310 |
"step": 30
|
| 311 |
},
|
| 312 |
{
|
| 313 |
+
"entropy": 1.7868350446224213,
|
| 314 |
+
"epoch": 0.9051094890510949,
|
| 315 |
+
"grad_norm": 3.71875,
|
| 316 |
+
"learning_rate": 2.7990381056766583e-05,
|
| 317 |
+
"loss": 1.4897,
|
| 318 |
+
"mean_token_accuracy": 0.5773478448390961,
|
| 319 |
+
"num_tokens": 61246.0,
|
| 320 |
"step": 31
|
| 321 |
},
|
| 322 |
{
|
| 323 |
+
"entropy": 1.8927763998508453,
|
| 324 |
+
"epoch": 0.9343065693430657,
|
| 325 |
+
"grad_norm": 5.03125,
|
| 326 |
+
"learning_rate": 2.772072144234639e-05,
|
| 327 |
+
"loss": 1.4658,
|
| 328 |
+
"mean_token_accuracy": 0.5965544059872627,
|
| 329 |
+
"num_tokens": 63057.0,
|
| 330 |
"step": 32
|
| 331 |
},
|
| 332 |
{
|
| 333 |
+
"entropy": 1.9243939369916916,
|
| 334 |
+
"epoch": 0.9635036496350365,
|
| 335 |
+
"grad_norm": 4.9375,
|
| 336 |
+
"learning_rate": 2.7435563588325627e-05,
|
| 337 |
+
"loss": 1.5646,
|
| 338 |
+
"mean_token_accuracy": 0.551388930529356,
|
| 339 |
+
"num_tokens": 64856.0,
|
| 340 |
"step": 33
|
| 341 |
},
|
| 342 |
{
|
| 343 |
+
"entropy": 1.945557788014412,
|
| 344 |
+
"epoch": 0.9927007299270073,
|
| 345 |
+
"grad_norm": 5.34375,
|
| 346 |
+
"learning_rate": 2.7135254915624213e-05,
|
| 347 |
+
"loss": 1.6558,
|
| 348 |
+
"mean_token_accuracy": 0.5641069300472736,
|
| 349 |
+
"num_tokens": 66564.0,
|
| 350 |
"step": 34
|
| 351 |
},
|
| 352 |
{
|
| 353 |
+
"entropy": 1.8289813995361328,
|
| 354 |
+
"epoch": 1.0,
|
| 355 |
+
"grad_norm": 12.5,
|
| 356 |
+
"learning_rate": 2.6820161304100828e-05,
|
| 357 |
+
"loss": 1.6743,
|
| 358 |
+
"mean_token_accuracy": 0.5590097606182098,
|
| 359 |
+
"num_tokens": 66897.0,
|
| 360 |
"step": 35
|
| 361 |
},
|
| 362 |
{
|
| 363 |
+
"entropy": 1.8240835815668106,
|
| 364 |
+
"epoch": 1.0291970802919708,
|
| 365 |
+
"grad_norm": 4.0,
|
| 366 |
+
"learning_rate": 2.649066664678467e-05,
|
| 367 |
+
"loss": 1.2519,
|
| 368 |
+
"mean_token_accuracy": 0.6510025560855865,
|
| 369 |
+
"num_tokens": 69125.0,
|
| 370 |
"step": 36
|
| 371 |
},
|
| 372 |
{
|
| 373 |
+
"entropy": 1.7388608753681183,
|
| 374 |
+
"epoch": 1.0583941605839415,
|
| 375 |
+
"grad_norm": 3.671875,
|
| 376 |
+
"learning_rate": 2.6147172382160913e-05,
|
| 377 |
+
"loss": 1.145,
|
| 378 |
+
"mean_token_accuracy": 0.6592915058135986,
|
| 379 |
+
"num_tokens": 71403.0,
|
| 380 |
"step": 37
|
| 381 |
},
|
| 382 |
{
|
| 383 |
+
"entropy": 1.7314125299453735,
|
| 384 |
+
"epoch": 1.0875912408759123,
|
| 385 |
+
"grad_norm": 3.84375,
|
| 386 |
+
"learning_rate": 2.5790097005079766e-05,
|
| 387 |
+
"loss": 1.2177,
|
| 388 |
+
"mean_token_accuracy": 0.6403542906045914,
|
| 389 |
+
"num_tokens": 73853.0,
|
| 390 |
"step": 38
|
| 391 |
},
|
| 392 |
{
|
| 393 |
+
"entropy": 1.9059295356273651,
|
| 394 |
+
"epoch": 1.1167883211678833,
|
| 395 |
+
"grad_norm": 5.09375,
|
| 396 |
+
"learning_rate": 2.541987555688496e-05,
|
| 397 |
+
"loss": 1.3537,
|
| 398 |
+
"mean_token_accuracy": 0.5938370451331139,
|
| 399 |
+
"num_tokens": 75484.0,
|
| 400 |
"step": 39
|
| 401 |
},
|
| 402 |
{
|
| 403 |
+
"entropy": 1.8351815044879913,
|
| 404 |
+
"epoch": 1.145985401459854,
|
| 405 |
+
"grad_norm": 5.03125,
|
| 406 |
+
"learning_rate": 2.5036959095382875e-05,
|
| 407 |
+
"loss": 1.1891,
|
| 408 |
+
"mean_token_accuracy": 0.6363263987004757,
|
| 409 |
+
"num_tokens": 77263.0,
|
| 410 |
"step": 40
|
| 411 |
},
|
| 412 |
{
|
| 413 |
+
"entropy": 1.856779396533966,
|
| 414 |
+
"epoch": 1.1751824817518248,
|
| 415 |
+
"grad_norm": 4.8125,
|
| 416 |
+
"learning_rate": 2.464181414529809e-05,
|
| 417 |
+
"loss": 1.3116,
|
| 418 |
+
"mean_token_accuracy": 0.625493511557579,
|
| 419 |
+
"num_tokens": 79113.0,
|
| 420 |
"step": 41
|
| 421 |
},
|
| 422 |
{
|
| 423 |
+
"entropy": 1.7603202909231186,
|
| 424 |
+
"epoch": 1.2043795620437956,
|
| 425 |
+
"grad_norm": 4.90625,
|
| 426 |
+
"learning_rate": 2.4234922129884873e-05,
|
| 427 |
+
"loss": 1.2056,
|
| 428 |
+
"mean_token_accuracy": 0.6308283284306526,
|
| 429 |
+
"num_tokens": 80962.0,
|
| 430 |
"step": 42
|
| 431 |
},
|
| 432 |
{
|
| 433 |
+
"entropy": 1.6366319358348846,
|
| 434 |
+
"epoch": 1.2335766423357664,
|
| 435 |
+
"grad_norm": 4.6875,
|
| 436 |
+
"learning_rate": 2.3816778784387097e-05,
|
| 437 |
+
"loss": 1.2438,
|
| 438 |
+
"mean_token_accuracy": 0.6533086150884628,
|
| 439 |
+
"num_tokens": 83095.0,
|
| 440 |
"step": 43
|
| 441 |
},
|
| 442 |
{
|
| 443 |
+
"entropy": 1.6320330947637558,
|
| 444 |
+
"epoch": 1.2627737226277373,
|
| 445 |
+
"grad_norm": 4.1875,
|
| 446 |
+
"learning_rate": 2.3387893552061202e-05,
|
| 447 |
+
"loss": 1.1647,
|
| 448 |
+
"mean_token_accuracy": 0.6589736789464951,
|
| 449 |
+
"num_tokens": 85383.0,
|
| 450 |
"step": 44
|
| 451 |
},
|
| 452 |
{
|
| 453 |
+
"entropy": 1.575496032834053,
|
| 454 |
+
"epoch": 1.2919708029197081,
|
| 455 |
+
"grad_norm": 4.65625,
|
| 456 |
+
"learning_rate": 2.2948788963498073e-05,
|
| 457 |
+
"loss": 1.1654,
|
| 458 |
+
"mean_token_accuracy": 0.6555850505828857,
|
| 459 |
+
"num_tokens": 87754.0,
|
| 460 |
"step": 45
|
| 461 |
},
|
| 462 |
{
|
| 463 |
+
"entropy": 1.64286208152771,
|
| 464 |
+
"epoch": 1.3211678832116789,
|
| 465 |
+
"grad_norm": 5.8125,
|
| 466 |
+
"learning_rate": 2.25e-05,
|
| 467 |
+
"loss": 1.3359,
|
| 468 |
+
"mean_token_accuracy": 0.649970181286335,
|
| 469 |
+
"num_tokens": 89289.0,
|
| 470 |
"step": 46
|
| 471 |
},
|
| 472 |
{
|
| 473 |
+
"entropy": 1.457002505660057,
|
| 474 |
+
"epoch": 1.3503649635036497,
|
| 475 |
+
"grad_norm": 4.75,
|
| 476 |
+
"learning_rate": 2.2042073441788363e-05,
|
| 477 |
+
"loss": 1.1513,
|
| 478 |
+
"mean_token_accuracy": 0.6784967109560966,
|
| 479 |
+
"num_tokens": 91666.0,
|
| 480 |
"step": 47
|
| 481 |
},
|
| 482 |
{
|
| 483 |
+
"entropy": 1.567281499505043,
|
| 484 |
+
"epoch": 1.3795620437956204,
|
| 485 |
+
"grad_norm": 6.78125,
|
| 486 |
+
"learning_rate": 2.157556720183616e-05,
|
| 487 |
+
"loss": 1.212,
|
| 488 |
+
"mean_token_accuracy": 0.6601979807019234,
|
| 489 |
+
"num_tokens": 93407.0,
|
| 490 |
"step": 48
|
| 491 |
},
|
| 492 |
{
|
| 493 |
+
"entropy": 1.4496354460716248,
|
| 494 |
+
"epoch": 1.4087591240875912,
|
| 495 |
+
"grad_norm": 4.90625,
|
| 496 |
+
"learning_rate": 2.1101049646137008e-05,
|
| 497 |
+
"loss": 1.074,
|
| 498 |
+
"mean_token_accuracy": 0.6734104976058006,
|
| 499 |
+
"num_tokens": 95819.0,
|
| 500 |
"step": 49
|
| 501 |
},
|
| 502 |
{
|
| 503 |
+
"entropy": 1.5027115792036057,
|
| 504 |
+
"epoch": 1.437956204379562,
|
| 505 |
+
"grad_norm": 4.65625,
|
| 506 |
+
"learning_rate": 2.0619098901238684e-05,
|
| 507 |
+
"loss": 1.1059,
|
| 508 |
+
"mean_token_accuracy": 0.6857927665114403,
|
| 509 |
+
"num_tokens": 98052.0,
|
| 510 |
"step": 50
|
| 511 |
},
|
| 512 |
{
|
| 513 |
+
"entropy": 1.5403490960597992,
|
| 514 |
+
"epoch": 1.4671532846715327,
|
| 515 |
+
"grad_norm": 5.75,
|
| 516 |
+
"learning_rate": 2.0130302149885033e-05,
|
| 517 |
+
"loss": 1.1573,
|
| 518 |
+
"mean_token_accuracy": 0.6808772906661034,
|
| 519 |
+
"num_tokens": 99865.0,
|
| 520 |
"step": 51
|
| 521 |
},
|
| 522 |
{
|
| 523 |
+
"entropy": 1.3851112127304077,
|
| 524 |
+
"epoch": 1.4963503649635037,
|
| 525 |
+
"grad_norm": 4.3125,
|
| 526 |
+
"learning_rate": 1.963525491562421e-05,
|
| 527 |
+
"loss": 1.0986,
|
| 528 |
+
"mean_token_accuracy": 0.669769361615181,
|
| 529 |
+
"num_tokens": 102444.0,
|
| 530 |
"step": 52
|
| 531 |
},
|
| 532 |
{
|
| 533 |
+
"entropy": 1.6086822748184204,
|
| 534 |
+
"epoch": 1.5255474452554745,
|
| 535 |
+
"grad_norm": 5.9375,
|
| 536 |
+
"learning_rate": 1.9134560337254986e-05,
|
| 537 |
+
"loss": 1.2058,
|
| 538 |
+
"mean_token_accuracy": 0.6342265903949738,
|
| 539 |
+
"num_tokens": 104135.0,
|
| 540 |
"step": 53
|
| 541 |
},
|
| 542 |
{
|
| 543 |
+
"entropy": 1.6186174154281616,
|
| 544 |
+
"epoch": 1.5547445255474452,
|
| 545 |
+
"grad_norm": 5.75,
|
| 546 |
+
"learning_rate": 1.8628828433995013e-05,
|
| 547 |
+
"loss": 1.1878,
|
| 548 |
+
"mean_token_accuracy": 0.6471928432583809,
|
| 549 |
+
"num_tokens": 105888.0,
|
| 550 |
"step": 54
|
| 551 |
},
|
| 552 |
{
|
| 553 |
+
"entropy": 1.636601522564888,
|
| 554 |
+
"epoch": 1.583941605839416,
|
| 555 |
+
"grad_norm": 6.40625,
|
| 556 |
+
"learning_rate": 1.8118675362266388e-05,
|
| 557 |
+
"loss": 1.2144,
|
| 558 |
+
"mean_token_accuracy": 0.669179767370224,
|
| 559 |
+
"num_tokens": 107324.0,
|
| 560 |
"step": 55
|
| 561 |
},
|
| 562 |
{
|
| 563 |
+
"entropy": 1.6150267571210861,
|
| 564 |
+
"epoch": 1.613138686131387,
|
| 565 |
+
"grad_norm": 6.21875,
|
| 566 |
+
"learning_rate": 1.760472266500396e-05,
|
| 567 |
+
"loss": 1.2551,
|
| 568 |
+
"mean_token_accuracy": 0.6627604365348816,
|
| 569 |
+
"num_tokens": 108844.0,
|
| 570 |
"step": 56
|
| 571 |
},
|
| 572 |
{
|
| 573 |
+
"entropy": 1.7444928288459778,
|
| 574 |
+
"epoch": 1.6423357664233578,
|
| 575 |
+
"grad_norm": 6.34375,
|
| 576 |
+
"learning_rate": 1.7087596514400982e-05,
|
| 577 |
+
"loss": 1.2656,
|
| 578 |
+
"mean_token_accuracy": 0.6279268711805344,
|
| 579 |
+
"num_tokens": 110263.0,
|
| 580 |
"step": 57
|
| 581 |
},
|
| 582 |
{
|
| 583 |
+
"entropy": 1.5423792004585266,
|
| 584 |
+
"epoch": 1.6715328467153285,
|
| 585 |
+
"grad_norm": 5.53125,
|
| 586 |
+
"learning_rate": 1.6567926949014805e-05,
|
| 587 |
+
"loss": 1.2103,
|
| 588 |
+
"mean_token_accuracy": 0.6224785149097443,
|
| 589 |
+
"num_tokens": 112199.0,
|
| 590 |
"step": 58
|
| 591 |
},
|
| 592 |
{
|
| 593 |
+
"entropy": 1.6031899452209473,
|
| 594 |
+
"epoch": 1.7007299270072993,
|
| 595 |
+
"grad_norm": 6.5,
|
| 596 |
+
"learning_rate": 1.604634710616188e-05,
|
| 597 |
+
"loss": 1.2274,
|
| 598 |
+
"mean_token_accuracy": 0.6428026333451271,
|
| 599 |
+
"num_tokens": 113911.0,
|
| 600 |
"step": 59
|
| 601 |
},
|
| 602 |
{
|
| 603 |
+
"entropy": 1.7055649012327194,
|
| 604 |
+
"epoch": 1.72992700729927,
|
| 605 |
+
"grad_norm": 6.6875,
|
| 606 |
+
"learning_rate": 1.552349245053752e-05,
|
| 607 |
+
"loss": 1.2889,
|
| 608 |
+
"mean_token_accuracy": 0.6419094651937485,
|
| 609 |
+
"num_tokens": 115316.0,
|
| 610 |
"step": 60
|
| 611 |
},
|
| 612 |
{
|
| 613 |
+
"entropy": 1.5212641060352325,
|
| 614 |
+
"epoch": 1.7591240875912408,
|
| 615 |
+
"grad_norm": 4.4375,
|
| 616 |
+
"learning_rate": 1.5e-05,
|
| 617 |
+
"loss": 1.0935,
|
| 618 |
+
"mean_token_accuracy": 0.6695626378059387,
|
| 619 |
+
"num_tokens": 118007.0,
|
| 620 |
"step": 61
|
| 621 |
},
|
| 622 |
{
|
| 623 |
+
"entropy": 1.781775563955307,
|
| 624 |
+
"epoch": 1.7883211678832116,
|
| 625 |
+
"grad_norm": 7.0,
|
| 626 |
+
"learning_rate": 1.447650754946249e-05,
|
| 627 |
+
"loss": 1.2709,
|
| 628 |
+
"mean_token_accuracy": 0.6656767651438713,
|
| 629 |
+
"num_tokens": 119232.0,
|
| 630 |
"step": 62
|
| 631 |
},
|
| 632 |
{
|
| 633 |
+
"entropy": 1.616694524884224,
|
| 634 |
+
"epoch": 1.8175182481751824,
|
| 635 |
+
"grad_norm": 6.3125,
|
| 636 |
+
"learning_rate": 1.3953652893838121e-05,
|
| 637 |
+
"loss": 1.2435,
|
| 638 |
+
"mean_token_accuracy": 0.6494908779859543,
|
| 639 |
+
"num_tokens": 120725.0,
|
| 640 |
"step": 63
|
| 641 |
},
|
| 642 |
{
|
| 643 |
+
"entropy": 1.7247931063175201,
|
| 644 |
+
"epoch": 1.8467153284671531,
|
| 645 |
+
"grad_norm": 7.15625,
|
| 646 |
+
"learning_rate": 1.3432073050985201e-05,
|
| 647 |
+
"loss": 1.3701,
|
| 648 |
+
"mean_token_accuracy": 0.6305030956864357,
|
| 649 |
+
"num_tokens": 122093.0,
|
| 650 |
"step": 64
|
| 651 |
},
|
| 652 |
{
|
| 653 |
+
"entropy": 1.590467780828476,
|
| 654 |
+
"epoch": 1.8759124087591241,
|
| 655 |
+
"grad_norm": 5.0,
|
| 656 |
+
"learning_rate": 1.2912403485599022e-05,
|
| 657 |
+
"loss": 1.263,
|
| 658 |
+
"mean_token_accuracy": 0.6583547666668892,
|
| 659 |
+
"num_tokens": 124333.0,
|
| 660 |
"step": 65
|
| 661 |
+
},
|
| 662 |
+
{
|
| 663 |
+
"entropy": 1.6301420778036118,
|
| 664 |
+
"epoch": 1.905109489051095,
|
| 665 |
+
"grad_norm": 5.3125,
|
| 666 |
+
"learning_rate": 1.2395277334996045e-05,
|
| 667 |
+
"loss": 1.1125,
|
| 668 |
+
"mean_token_accuracy": 0.650074191391468,
|
| 669 |
+
"num_tokens": 126272.0,
|
| 670 |
+
"step": 66
|
| 671 |
+
},
|
| 672 |
+
{
|
| 673 |
+
"entropy": 1.5050681680440903,
|
| 674 |
+
"epoch": 1.9343065693430657,
|
| 675 |
+
"grad_norm": 4.28125,
|
| 676 |
+
"learning_rate": 1.1881324637733613e-05,
|
| 677 |
+
"loss": 1.037,
|
| 678 |
+
"mean_token_accuracy": 0.6733650118112564,
|
| 679 |
+
"num_tokens": 128615.0,
|
| 680 |
+
"step": 67
|
| 681 |
+
},
|
| 682 |
+
{
|
| 683 |
+
"entropy": 1.5582159608602524,
|
| 684 |
+
"epoch": 1.9635036496350367,
|
| 685 |
+
"grad_norm": 4.34375,
|
| 686 |
+
"learning_rate": 1.1371171566004986e-05,
|
| 687 |
+
"loss": 1.0951,
|
| 688 |
+
"mean_token_accuracy": 0.6506948918104172,
|
| 689 |
+
"num_tokens": 131279.0,
|
| 690 |
+
"step": 68
|
| 691 |
+
},
|
| 692 |
+
{
|
| 693 |
+
"entropy": 1.6561681628227234,
|
| 694 |
+
"epoch": 1.9927007299270074,
|
| 695 |
+
"grad_norm": 5.75,
|
| 696 |
+
"learning_rate": 1.0865439662745013e-05,
|
| 697 |
+
"loss": 1.1486,
|
| 698 |
+
"mean_token_accuracy": 0.6755311414599419,
|
| 699 |
+
"num_tokens": 132847.0,
|
| 700 |
+
"step": 69
|
| 701 |
+
},
|
| 702 |
+
{
|
| 703 |
+
"entropy": 1.4383031129837036,
|
| 704 |
+
"epoch": 2.0,
|
| 705 |
+
"grad_norm": 7.4375,
|
| 706 |
+
"learning_rate": 1.036474508437579e-05,
|
| 707 |
+
"loss": 1.1032,
|
| 708 |
+
"mean_token_accuracy": 0.6792386174201965,
|
| 709 |
+
"num_tokens": 133794.0,
|
| 710 |
+
"step": 70
|
| 711 |
+
},
|
| 712 |
+
{
|
| 713 |
+
"entropy": 1.5033023059368134,
|
| 714 |
+
"epoch": 2.0291970802919708,
|
| 715 |
+
"grad_norm": 4.09375,
|
| 716 |
+
"learning_rate": 9.86969785011497e-06,
|
| 717 |
+
"loss": 0.8414,
|
| 718 |
+
"mean_token_accuracy": 0.7257160544395447,
|
| 719 |
+
"num_tokens": 135994.0,
|
| 720 |
+
"step": 71
|
| 721 |
+
},
|
| 722 |
+
{
|
| 723 |
+
"entropy": 1.588482990860939,
|
| 724 |
+
"epoch": 2.0583941605839415,
|
| 725 |
+
"grad_norm": 5.4375,
|
| 726 |
+
"learning_rate": 9.380901098761319e-06,
|
| 727 |
+
"loss": 0.8667,
|
| 728 |
+
"mean_token_accuracy": 0.7469649091362953,
|
| 729 |
+
"num_tokens": 137554.0,
|
| 730 |
+
"step": 72
|
| 731 |
+
},
|
| 732 |
+
{
|
| 733 |
+
"entropy": 1.539756417274475,
|
| 734 |
+
"epoch": 2.0875912408759123,
|
| 735 |
+
"grad_norm": 4.3125,
|
| 736 |
+
"learning_rate": 8.898950353863e-06,
|
| 737 |
+
"loss": 0.8192,
|
| 738 |
+
"mean_token_accuracy": 0.7514503225684166,
|
| 739 |
+
"num_tokens": 139542.0,
|
| 740 |
+
"step": 73
|
| 741 |
+
},
|
| 742 |
+
{
|
| 743 |
+
"entropy": 1.5114945620298386,
|
| 744 |
+
"epoch": 2.116788321167883,
|
| 745 |
+
"grad_norm": 4.25,
|
| 746 |
+
"learning_rate": 8.424432798163838e-06,
|
| 747 |
+
"loss": 0.9041,
|
| 748 |
+
"mean_token_accuracy": 0.7257768511772156,
|
| 749 |
+
"num_tokens": 141721.0,
|
| 750 |
+
"step": 74
|
| 751 |
+
},
|
| 752 |
+
{
|
| 753 |
+
"entropy": 1.4715029448270798,
|
| 754 |
+
"epoch": 2.145985401459854,
|
| 755 |
+
"grad_norm": 4.375,
|
| 756 |
+
"learning_rate": 7.957926558211643e-06,
|
| 757 |
+
"loss": 0.8884,
|
| 758 |
+
"mean_token_accuracy": 0.7411475032567978,
|
| 759 |
+
"num_tokens": 143837.0,
|
| 760 |
+
"step": 75
|
| 761 |
+
},
|
| 762 |
+
{
|
| 763 |
+
"entropy": 1.375910922884941,
|
| 764 |
+
"epoch": 2.1751824817518246,
|
| 765 |
+
"grad_norm": 4.0625,
|
| 766 |
+
"learning_rate": 7.500000000000004e-06,
|
| 767 |
+
"loss": 0.8403,
|
| 768 |
+
"mean_token_accuracy": 0.7337475717067719,
|
| 769 |
+
"num_tokens": 146069.0,
|
| 770 |
+
"step": 76
|
| 771 |
+
},
|
| 772 |
+
{
|
| 773 |
+
"entropy": 1.530395969748497,
|
| 774 |
+
"epoch": 2.204379562043796,
|
| 775 |
+
"grad_norm": 4.8125,
|
| 776 |
+
"learning_rate": 7.051211036501928e-06,
|
| 777 |
+
"loss": 0.9023,
|
| 778 |
+
"mean_token_accuracy": 0.7458862364292145,
|
| 779 |
+
"num_tokens": 147948.0,
|
| 780 |
+
"step": 77
|
| 781 |
+
},
|
| 782 |
+
{
|
| 783 |
+
"entropy": 1.5619382560253143,
|
| 784 |
+
"epoch": 2.2335766423357666,
|
| 785 |
+
"grad_norm": 5.375,
|
| 786 |
+
"learning_rate": 6.6121064479388e-06,
|
| 787 |
+
"loss": 0.9471,
|
| 788 |
+
"mean_token_accuracy": 0.7247473746538162,
|
| 789 |
+
"num_tokens": 149664.0,
|
| 790 |
+
"step": 78
|
| 791 |
+
},
|
| 792 |
+
{
|
| 793 |
+
"entropy": 1.4002738296985626,
|
| 794 |
+
"epoch": 2.2627737226277373,
|
| 795 |
+
"grad_norm": 4.90625,
|
| 796 |
+
"learning_rate": 6.1832212156129045e-06,
|
| 797 |
+
"loss": 0.8002,
|
| 798 |
+
"mean_token_accuracy": 0.7359691336750984,
|
| 799 |
+
"num_tokens": 151422.0,
|
| 800 |
+
"step": 79
|
| 801 |
+
},
|
| 802 |
+
{
|
| 803 |
+
"entropy": 1.3783821165561676,
|
| 804 |
+
"epoch": 2.291970802919708,
|
| 805 |
+
"grad_norm": 4.875,
|
| 806 |
+
"learning_rate": 5.765077870115126e-06,
|
| 807 |
+
"loss": 0.9352,
|
| 808 |
+
"mean_token_accuracy": 0.7229901030659676,
|
| 809 |
+
"num_tokens": 153330.0,
|
| 810 |
+
"step": 80
|
| 811 |
+
},
|
| 812 |
+
{
|
| 813 |
+
"entropy": 1.3214146196842194,
|
| 814 |
+
"epoch": 2.321167883211679,
|
| 815 |
+
"grad_norm": 4.875,
|
| 816 |
+
"learning_rate": 5.3581858547019095e-06,
|
| 817 |
+
"loss": 0.7626,
|
| 818 |
+
"mean_token_accuracy": 0.7818252220749855,
|
| 819 |
+
"num_tokens": 155088.0,
|
| 820 |
+
"step": 81
|
| 821 |
+
},
|
| 822 |
+
{
|
| 823 |
+
"entropy": 1.2702767699956894,
|
| 824 |
+
"epoch": 2.3503649635036497,
|
| 825 |
+
"grad_norm": 4.375,
|
| 826 |
+
"learning_rate": 4.963040904617131e-06,
|
| 827 |
+
"loss": 0.7893,
|
| 828 |
+
"mean_token_accuracy": 0.7699355036020279,
|
| 829 |
+
"num_tokens": 157396.0,
|
| 830 |
+
"step": 82
|
| 831 |
+
},
|
| 832 |
+
{
|
| 833 |
+
"entropy": 1.397829994559288,
|
| 834 |
+
"epoch": 2.3795620437956204,
|
| 835 |
+
"grad_norm": 5.25,
|
| 836 |
+
"learning_rate": 4.58012444311504e-06,
|
| 837 |
+
"loss": 0.9191,
|
| 838 |
+
"mean_token_accuracy": 0.7331462875008583,
|
| 839 |
+
"num_tokens": 159218.0,
|
| 840 |
+
"step": 83
|
| 841 |
+
},
|
| 842 |
+
{
|
| 843 |
+
"entropy": 1.2017180174589157,
|
| 844 |
+
"epoch": 2.408759124087591,
|
| 845 |
+
"grad_norm": 3.6875,
|
| 846 |
+
"learning_rate": 4.209902994920236e-06,
|
| 847 |
+
"loss": 0.8082,
|
| 848 |
+
"mean_token_accuracy": 0.7587887346744537,
|
| 849 |
+
"num_tokens": 162386.0,
|
| 850 |
+
"step": 84
|
| 851 |
+
},
|
| 852 |
+
{
|
| 853 |
+
"entropy": 1.374891072511673,
|
| 854 |
+
"epoch": 2.437956204379562,
|
| 855 |
+
"grad_norm": 5.09375,
|
| 856 |
+
"learning_rate": 3.852827617839085e-06,
|
| 857 |
+
"loss": 0.8665,
|
| 858 |
+
"mean_token_accuracy": 0.7603413909673691,
|
| 859 |
+
"num_tokens": 164138.0,
|
| 860 |
+
"step": 85
|
| 861 |
+
},
|
| 862 |
+
{
|
| 863 |
+
"entropy": 1.3341291099786758,
|
| 864 |
+
"epoch": 2.4671532846715327,
|
| 865 |
+
"grad_norm": 4.6875,
|
| 866 |
+
"learning_rate": 3.5093333532153316e-06,
|
| 867 |
+
"loss": 0.8604,
|
| 868 |
+
"mean_token_accuracy": 0.7294721901416779,
|
| 869 |
+
"num_tokens": 166308.0,
|
| 870 |
+
"step": 86
|
| 871 |
+
},
|
| 872 |
+
{
|
| 873 |
+
"entropy": 1.3214628398418427,
|
| 874 |
+
"epoch": 2.4963503649635035,
|
| 875 |
+
"grad_norm": 5.4375,
|
| 876 |
+
"learning_rate": 3.1798386958991715e-06,
|
| 877 |
+
"loss": 0.8978,
|
| 878 |
+
"mean_token_accuracy": 0.7371588498353958,
|
| 879 |
+
"num_tokens": 168073.0,
|
| 880 |
+
"step": 87
|
| 881 |
+
},
|
| 882 |
+
{
|
| 883 |
+
"entropy": 1.358703538775444,
|
| 884 |
+
"epoch": 2.5255474452554747,
|
| 885 |
+
"grad_norm": 5.125,
|
| 886 |
+
"learning_rate": 2.86474508437579e-06,
|
| 887 |
+
"loss": 0.859,
|
| 888 |
+
"mean_token_accuracy": 0.7255095988512039,
|
| 889 |
+
"num_tokens": 169979.0,
|
| 890 |
+
"step": 88
|
| 891 |
+
},
|
| 892 |
+
{
|
| 893 |
+
"entropy": 1.258324310183525,
|
| 894 |
+
"epoch": 2.554744525547445,
|
| 895 |
+
"grad_norm": 4.15625,
|
| 896 |
+
"learning_rate": 2.564436411674376e-06,
|
| 897 |
+
"loss": 0.825,
|
| 898 |
+
"mean_token_accuracy": 0.7614458128809929,
|
| 899 |
+
"num_tokens": 172706.0,
|
| 900 |
+
"step": 89
|
| 901 |
+
},
|
| 902 |
+
{
|
| 903 |
+
"entropy": 1.329784169793129,
|
| 904 |
+
"epoch": 2.5839416058394162,
|
| 905 |
+
"grad_norm": 5.40625,
|
| 906 |
+
"learning_rate": 2.279278557653611e-06,
|
| 907 |
+
"loss": 0.8799,
|
| 908 |
+
"mean_token_accuracy": 0.7584780603647232,
|
| 909 |
+
"num_tokens": 174586.0,
|
| 910 |
+
"step": 90
|
| 911 |
+
},
|
| 912 |
+
{
|
| 913 |
+
"entropy": 1.2622641026973724,
|
| 914 |
+
"epoch": 2.613138686131387,
|
| 915 |
+
"grad_norm": 5.125,
|
| 916 |
+
"learning_rate": 2.0096189432334194e-06,
|
| 917 |
+
"loss": 0.8348,
|
| 918 |
+
"mean_token_accuracy": 0.7513260990381241,
|
| 919 |
+
"num_tokens": 176525.0,
|
| 920 |
+
"step": 91
|
| 921 |
+
},
|
| 922 |
+
{
|
| 923 |
+
"entropy": 1.2846813797950745,
|
| 924 |
+
"epoch": 2.6423357664233578,
|
| 925 |
+
"grad_norm": 5.0,
|
| 926 |
+
"learning_rate": 1.7557861071160953e-06,
|
| 927 |
+
"loss": 0.7697,
|
| 928 |
+
"mean_token_accuracy": 0.7566402554512024,
|
| 929 |
+
"num_tokens": 178535.0,
|
| 930 |
+
"step": 92
|
| 931 |
+
},
|
| 932 |
+
{
|
| 933 |
+
"entropy": 1.2429047673940659,
|
| 934 |
+
"epoch": 2.6715328467153285,
|
| 935 |
+
"grad_norm": 4.1875,
|
| 936 |
+
"learning_rate": 1.518089305512498e-06,
|
| 937 |
+
"loss": 0.8523,
|
| 938 |
+
"mean_token_accuracy": 0.7609995678067207,
|
| 939 |
+
"num_tokens": 181688.0,
|
| 940 |
+
"step": 93
|
| 941 |
+
},
|
| 942 |
+
{
|
| 943 |
+
"entropy": 1.2306764125823975,
|
| 944 |
+
"epoch": 2.7007299270072993,
|
| 945 |
+
"grad_norm": 5.6875,
|
| 946 |
+
"learning_rate": 1.2968181353609854e-06,
|
| 947 |
+
"loss": 0.795,
|
| 948 |
+
"mean_token_accuracy": 0.7538608759641647,
|
| 949 |
+
"num_tokens": 183350.0,
|
| 950 |
+
"step": 94
|
| 951 |
+
},
|
| 952 |
+
{
|
| 953 |
+
"entropy": 1.2729838192462921,
|
| 954 |
+
"epoch": 2.72992700729927,
|
| 955 |
+
"grad_norm": 5.25,
|
| 956 |
+
"learning_rate": 1.0922421814981904e-06,
|
| 957 |
+
"loss": 0.8463,
|
| 958 |
+
"mean_token_accuracy": 0.7443541586399078,
|
| 959 |
+
"num_tokens": 185369.0,
|
| 960 |
+
"step": 95
|
| 961 |
+
},
|
| 962 |
+
{
|
| 963 |
+
"entropy": 1.2911252602934837,
|
| 964 |
+
"epoch": 2.759124087591241,
|
| 965 |
+
"grad_norm": 5.125,
|
| 966 |
+
"learning_rate": 9.046106882113753e-07,
|
| 967 |
+
"loss": 0.7471,
|
| 968 |
+
"mean_token_accuracy": 0.752311646938324,
|
| 969 |
+
"num_tokens": 187493.0,
|
| 970 |
+
"step": 96
|
| 971 |
+
},
|
| 972 |
+
{
|
| 973 |
+
"entropy": 1.28748519718647,
|
| 974 |
+
"epoch": 2.7883211678832116,
|
| 975 |
+
"grad_norm": 6.4375,
|
| 976 |
+
"learning_rate": 7.341522555726971e-07,
|
| 977 |
+
"loss": 0.7536,
|
| 978 |
+
"mean_token_accuracy": 0.7757409885525703,
|
| 979 |
+
"num_tokens": 188864.0,
|
| 980 |
+
"step": 97
|
| 981 |
+
},
|
| 982 |
+
{
|
| 983 |
+
"entropy": 1.2816387563943863,
|
| 984 |
+
"epoch": 2.8175182481751824,
|
| 985 |
+
"grad_norm": 5.46875,
|
| 986 |
+
"learning_rate": 5.810745609252166e-07,
|
| 987 |
+
"loss": 0.9127,
|
| 988 |
+
"mean_token_accuracy": 0.7290580719709396,
|
| 989 |
+
"num_tokens": 190843.0,
|
| 990 |
+
"step": 98
|
| 991 |
+
},
|
| 992 |
+
{
|
| 993 |
+
"entropy": 1.4024466425180435,
|
| 994 |
+
"epoch": 2.846715328467153,
|
| 995 |
+
"grad_norm": 6.71875,
|
| 996 |
+
"learning_rate": 4.455641058600529e-07,
|
| 997 |
+
"loss": 0.9032,
|
| 998 |
+
"mean_token_accuracy": 0.7520110681653023,
|
| 999 |
+
"num_tokens": 192230.0,
|
| 1000 |
+
"step": 99
|
| 1001 |
+
},
|
| 1002 |
+
{
|
| 1003 |
+
"entropy": 1.354932889342308,
|
| 1004 |
+
"epoch": 2.875912408759124,
|
| 1005 |
+
"grad_norm": 6.71875,
|
| 1006 |
+
"learning_rate": 3.277859889929147e-07,
|
| 1007 |
+
"loss": 0.7987,
|
| 1008 |
+
"mean_token_accuracy": 0.785490907728672,
|
| 1009 |
+
"num_tokens": 193518.0,
|
| 1010 |
+
"step": 100
|
| 1011 |
}
|
| 1012 |
],
|
| 1013 |
"logging_steps": 1,
|
| 1014 |
+
"max_steps": 105,
|
| 1015 |
"num_input_tokens_seen": 0,
|
| 1016 |
"num_train_epochs": 3,
|
| 1017 |
"save_steps": 5,
|
|
|
|
| 1027 |
"attributes": {}
|
| 1028 |
}
|
| 1029 |
},
|
| 1030 |
+
"total_flos": 5186447183892480.0,
|
| 1031 |
"train_batch_size": 2,
|
| 1032 |
"trial_name": null,
|
| 1033 |
"trial_params": null
|
training_args.bin
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 6353
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f89ff4081cf45cebdd9100f3809aacae74d1773ff8c2b672defb6af57e4a514c
|
| 3 |
size 6353
|