End of training
Browse files- README.md +2 -1
- all_results.json +6 -6
- train_results.json +6 -6
- trainer_state.json +419 -440
- training_loss.png +0 -0
README.md
CHANGED
|
@@ -4,6 +4,7 @@ license: apache-2.0
|
|
| 4 |
base_model: Qwen/Qwen2.5-7B-Instruct
|
| 5 |
tags:
|
| 6 |
- llama-factory
|
|
|
|
| 7 |
- generated_from_trainer
|
| 8 |
model-index:
|
| 9 |
- name: llama3-1_8b_4o_annotated_aime
|
|
@@ -15,7 +16,7 @@ should probably proofread and complete it, then remove this comment. -->
|
|
| 15 |
|
| 16 |
# llama3-1_8b_4o_annotated_aime
|
| 17 |
|
| 18 |
-
This model is a fine-tuned version of [Qwen/Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) on
|
| 19 |
|
| 20 |
## Model description
|
| 21 |
|
|
|
|
| 4 |
base_model: Qwen/Qwen2.5-7B-Instruct
|
| 5 |
tags:
|
| 6 |
- llama-factory
|
| 7 |
+
- full
|
| 8 |
- generated_from_trainer
|
| 9 |
model-index:
|
| 10 |
- name: llama3-1_8b_4o_annotated_aime
|
|
|
|
| 16 |
|
| 17 |
# llama3-1_8b_4o_annotated_aime
|
| 18 |
|
| 19 |
+
This model is a fine-tuned version of [Qwen/Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) on the mlfoundations-dev/4o_annotated_aime dataset.
|
| 20 |
|
| 21 |
## Model description
|
| 22 |
|
all_results.json
CHANGED
|
@@ -1,8 +1,8 @@
|
|
| 1 |
{
|
| 2 |
-
"epoch":
|
| 3 |
-
"total_flos":
|
| 4 |
-
"train_loss": 0.
|
| 5 |
-
"train_runtime":
|
| 6 |
-
"train_samples_per_second":
|
| 7 |
-
"train_steps_per_second": 0.
|
| 8 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"epoch": 2.9626168224299065,
|
| 3 |
+
"total_flos": 18524510396416.0,
|
| 4 |
+
"train_loss": 0.47221575805119104,
|
| 5 |
+
"train_runtime": 641.5687,
|
| 6 |
+
"train_samples_per_second": 15.913,
|
| 7 |
+
"train_steps_per_second": 0.164
|
| 8 |
}
|
train_results.json
CHANGED
|
@@ -1,8 +1,8 @@
|
|
| 1 |
{
|
| 2 |
-
"epoch":
|
| 3 |
-
"total_flos":
|
| 4 |
-
"train_loss": 0.
|
| 5 |
-
"train_runtime":
|
| 6 |
-
"train_samples_per_second":
|
| 7 |
-
"train_steps_per_second": 0.
|
| 8 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"epoch": 2.9626168224299065,
|
| 3 |
+
"total_flos": 18524510396416.0,
|
| 4 |
+
"train_loss": 0.47221575805119104,
|
| 5 |
+
"train_runtime": 641.5687,
|
| 6 |
+
"train_samples_per_second": 15.913,
|
| 7 |
+
"train_steps_per_second": 0.164
|
| 8 |
}
|
trainer_state.json
CHANGED
|
@@ -1,781 +1,760 @@
|
|
| 1 |
{
|
| 2 |
"best_metric": null,
|
| 3 |
"best_model_checkpoint": null,
|
| 4 |
-
"epoch":
|
| 5 |
"eval_steps": 500,
|
| 6 |
-
"global_step":
|
| 7 |
"is_hyper_param_search": false,
|
| 8 |
"is_local_process_zero": true,
|
| 9 |
"is_world_process_zero": true,
|
| 10 |
"log_history": [
|
| 11 |
{
|
| 12 |
-
"epoch": 0.
|
| 13 |
-
"grad_norm": 5.
|
| 14 |
"learning_rate": 9.090909090909091e-07,
|
| 15 |
-
"loss": 0.
|
| 16 |
"step": 1
|
| 17 |
},
|
| 18 |
{
|
| 19 |
-
"epoch": 0.
|
| 20 |
-
"grad_norm": 5.
|
| 21 |
"learning_rate": 1.8181818181818183e-06,
|
| 22 |
-
"loss": 0.
|
| 23 |
"step": 2
|
| 24 |
},
|
| 25 |
{
|
| 26 |
-
"epoch": 0.
|
| 27 |
-
"grad_norm": 5.
|
| 28 |
"learning_rate": 2.7272727272727272e-06,
|
| 29 |
-
"loss": 0.
|
| 30 |
"step": 3
|
| 31 |
},
|
| 32 |
{
|
| 33 |
-
"epoch": 0.
|
| 34 |
-
"grad_norm": 5.
|
| 35 |
"learning_rate": 3.6363636363636366e-06,
|
| 36 |
-
"loss": 0.
|
| 37 |
"step": 4
|
| 38 |
},
|
| 39 |
{
|
| 40 |
-
"epoch": 0.
|
| 41 |
-
"grad_norm": 3.
|
| 42 |
"learning_rate": 4.5454545454545455e-06,
|
| 43 |
-
"loss": 0.
|
| 44 |
"step": 5
|
| 45 |
},
|
| 46 |
{
|
| 47 |
-
"epoch": 0.
|
| 48 |
-
"grad_norm": 2.
|
| 49 |
"learning_rate": 5.4545454545454545e-06,
|
| 50 |
-
"loss": 0.
|
| 51 |
"step": 6
|
| 52 |
},
|
| 53 |
{
|
| 54 |
-
"epoch": 0.
|
| 55 |
-
"grad_norm":
|
| 56 |
"learning_rate": 6.363636363636364e-06,
|
| 57 |
-
"loss": 0.
|
| 58 |
"step": 7
|
| 59 |
},
|
| 60 |
{
|
| 61 |
-
"epoch": 0.
|
| 62 |
-
"grad_norm": 1.
|
| 63 |
"learning_rate": 7.272727272727273e-06,
|
| 64 |
"loss": 0.5054,
|
| 65 |
"step": 8
|
| 66 |
},
|
| 67 |
{
|
| 68 |
-
"epoch": 0.
|
| 69 |
-
"grad_norm": 1.
|
| 70 |
"learning_rate": 8.181818181818183e-06,
|
| 71 |
-
"loss": 0.
|
| 72 |
"step": 9
|
| 73 |
},
|
| 74 |
{
|
| 75 |
-
"epoch": 0.
|
| 76 |
-
"grad_norm": 1.
|
| 77 |
"learning_rate": 9.090909090909091e-06,
|
| 78 |
"loss": 0.5433,
|
| 79 |
"step": 10
|
| 80 |
},
|
| 81 |
{
|
| 82 |
-
"epoch": 0.
|
| 83 |
-
"grad_norm": 1.
|
| 84 |
"learning_rate": 1e-05,
|
| 85 |
-
"loss": 0.
|
| 86 |
"step": 11
|
| 87 |
},
|
| 88 |
{
|
| 89 |
-
"epoch": 0.
|
| 90 |
-
"grad_norm": 1.
|
| 91 |
-
"learning_rate": 9.
|
| 92 |
-
"loss": 0.
|
| 93 |
"step": 12
|
| 94 |
},
|
| 95 |
{
|
| 96 |
-
"epoch": 0.
|
| 97 |
-
"grad_norm": 1.
|
| 98 |
-
"learning_rate": 9.
|
| 99 |
-
"loss": 0.
|
| 100 |
"step": 13
|
| 101 |
},
|
| 102 |
{
|
| 103 |
-
"epoch": 0.
|
| 104 |
-
"grad_norm": 1.
|
| 105 |
-
"learning_rate": 9.
|
| 106 |
-
"loss": 0.
|
| 107 |
"step": 14
|
| 108 |
},
|
| 109 |
{
|
| 110 |
-
"epoch": 0.
|
| 111 |
-
"grad_norm": 1.
|
| 112 |
-
"learning_rate": 9.
|
| 113 |
-
"loss": 0.
|
| 114 |
"step": 15
|
| 115 |
},
|
| 116 |
{
|
| 117 |
-
"epoch": 0.
|
| 118 |
-
"grad_norm": 1.
|
| 119 |
-
"learning_rate": 9.
|
| 120 |
-
"loss": 0.
|
| 121 |
"step": 16
|
| 122 |
},
|
| 123 |
{
|
| 124 |
-
"epoch": 0.
|
| 125 |
-
"grad_norm": 1.
|
| 126 |
-
"learning_rate": 9.
|
| 127 |
-
"loss": 0.
|
| 128 |
"step": 17
|
| 129 |
},
|
| 130 |
{
|
| 131 |
-
"epoch": 0.
|
| 132 |
-
"grad_norm": 0.
|
| 133 |
-
"learning_rate": 9.
|
| 134 |
-
"loss": 0.
|
| 135 |
"step": 18
|
| 136 |
},
|
| 137 |
{
|
| 138 |
-
"epoch": 0.
|
| 139 |
-
"grad_norm":
|
| 140 |
-
"learning_rate": 9.
|
| 141 |
-
"loss": 0.
|
| 142 |
"step": 19
|
| 143 |
},
|
| 144 |
{
|
| 145 |
-
"epoch": 0.
|
| 146 |
-
"grad_norm": 0.
|
| 147 |
-
"learning_rate": 9.
|
| 148 |
-
"loss": 0.
|
| 149 |
"step": 20
|
| 150 |
},
|
| 151 |
{
|
| 152 |
-
"epoch": 0.
|
| 153 |
-
"grad_norm": 0.
|
| 154 |
-
"learning_rate": 9.
|
| 155 |
-
"loss": 0.
|
| 156 |
"step": 21
|
| 157 |
},
|
| 158 |
{
|
| 159 |
-
"epoch": 0.
|
| 160 |
-
"grad_norm": 0.
|
| 161 |
-
"learning_rate": 9.
|
| 162 |
-
"loss": 0.
|
| 163 |
"step": 22
|
| 164 |
},
|
| 165 |
{
|
| 166 |
-
"epoch": 0.
|
| 167 |
-
"grad_norm": 0.
|
| 168 |
-
"learning_rate": 9.
|
| 169 |
-
"loss": 0.
|
| 170 |
"step": 23
|
| 171 |
},
|
| 172 |
{
|
| 173 |
-
"epoch": 0.
|
| 174 |
-
"grad_norm": 0.
|
| 175 |
-
"learning_rate": 9.
|
| 176 |
-
"loss": 0.
|
| 177 |
"step": 24
|
| 178 |
},
|
| 179 |
{
|
| 180 |
-
"epoch": 0.
|
| 181 |
-
"grad_norm": 0.
|
| 182 |
-
"learning_rate": 9.
|
| 183 |
-
"loss": 0.
|
| 184 |
"step": 25
|
| 185 |
},
|
| 186 |
{
|
| 187 |
-
"epoch": 0.
|
| 188 |
-
"grad_norm": 0.
|
| 189 |
-
"learning_rate": 9.
|
| 190 |
-
"loss": 0.
|
| 191 |
"step": 26
|
| 192 |
},
|
| 193 |
{
|
| 194 |
-
"epoch": 0.
|
| 195 |
-
"grad_norm": 0.
|
| 196 |
-
"learning_rate": 9.
|
| 197 |
-
"loss": 0.
|
| 198 |
"step": 27
|
| 199 |
},
|
| 200 |
{
|
| 201 |
-
"epoch": 0.
|
| 202 |
-
"grad_norm": 0.
|
| 203 |
-
"learning_rate": 9.
|
| 204 |
-
"loss": 0.
|
| 205 |
"step": 28
|
| 206 |
},
|
| 207 |
{
|
| 208 |
-
"epoch": 0.
|
| 209 |
-
"grad_norm": 0.
|
| 210 |
-
"learning_rate": 9.
|
| 211 |
-
"loss": 0.
|
| 212 |
"step": 29
|
| 213 |
},
|
| 214 |
{
|
| 215 |
-
"epoch": 0.
|
| 216 |
-
"grad_norm": 1.
|
| 217 |
-
"learning_rate": 9.
|
| 218 |
-
"loss": 0.
|
| 219 |
"step": 30
|
| 220 |
},
|
| 221 |
{
|
| 222 |
-
"epoch": 0.
|
| 223 |
-
"grad_norm": 0.
|
| 224 |
-
"learning_rate": 8.
|
| 225 |
-
"loss": 0.
|
| 226 |
"step": 31
|
| 227 |
},
|
| 228 |
{
|
| 229 |
-
"epoch": 0.
|
| 230 |
-
"grad_norm": 0.
|
| 231 |
-
"learning_rate": 8.
|
| 232 |
-
"loss": 0.
|
| 233 |
"step": 32
|
| 234 |
},
|
| 235 |
{
|
| 236 |
-
"epoch": 0.
|
| 237 |
-
"grad_norm": 0.
|
| 238 |
-
"learning_rate": 8.
|
| 239 |
-
"loss": 0.
|
| 240 |
"step": 33
|
| 241 |
},
|
| 242 |
{
|
| 243 |
-
"epoch": 0.
|
| 244 |
-
"grad_norm": 0.
|
| 245 |
-
"learning_rate": 8.
|
| 246 |
-
"loss": 0.
|
| 247 |
"step": 34
|
| 248 |
},
|
| 249 |
{
|
| 250 |
-
"epoch": 0.
|
| 251 |
-
"grad_norm": 0.
|
| 252 |
-
"learning_rate": 8.
|
| 253 |
-
"loss": 0.
|
| 254 |
"step": 35
|
| 255 |
},
|
| 256 |
{
|
| 257 |
-
"epoch": 1.
|
| 258 |
-
"grad_norm":
|
| 259 |
-
"learning_rate": 8.
|
| 260 |
-
"loss": 0.
|
| 261 |
"step": 36
|
| 262 |
},
|
| 263 |
{
|
| 264 |
-
"epoch": 1.
|
| 265 |
-
"grad_norm":
|
| 266 |
-
"learning_rate": 8.
|
| 267 |
-
"loss": 0.
|
| 268 |
"step": 37
|
| 269 |
},
|
| 270 |
{
|
| 271 |
-
"epoch": 1.
|
| 272 |
-
"grad_norm":
|
| 273 |
-
"learning_rate": 8.
|
| 274 |
-
"loss": 0.
|
| 275 |
"step": 38
|
| 276 |
},
|
| 277 |
{
|
| 278 |
-
"epoch": 1.
|
| 279 |
-
"grad_norm": 0.
|
| 280 |
-
"learning_rate":
|
| 281 |
-
"loss": 0.
|
| 282 |
"step": 39
|
| 283 |
},
|
| 284 |
{
|
| 285 |
-
"epoch": 1.
|
| 286 |
-
"grad_norm": 0.
|
| 287 |
-
"learning_rate": 7.
|
| 288 |
-
"loss": 0.
|
| 289 |
"step": 40
|
| 290 |
},
|
| 291 |
{
|
| 292 |
-
"epoch": 1.
|
| 293 |
-
"grad_norm": 0.
|
| 294 |
-
"learning_rate": 7.
|
| 295 |
-
"loss": 0.
|
| 296 |
"step": 41
|
| 297 |
},
|
| 298 |
{
|
| 299 |
-
"epoch": 1.
|
| 300 |
-
"grad_norm": 0.
|
| 301 |
-
"learning_rate": 7.
|
| 302 |
-
"loss": 0.
|
| 303 |
"step": 42
|
| 304 |
},
|
| 305 |
{
|
| 306 |
-
"epoch": 1.
|
| 307 |
-
"grad_norm": 0.
|
| 308 |
-
"learning_rate": 7.
|
| 309 |
-
"loss": 0.
|
| 310 |
"step": 43
|
| 311 |
},
|
| 312 |
{
|
| 313 |
-
"epoch": 1.
|
| 314 |
-
"grad_norm": 0.
|
| 315 |
-
"learning_rate": 7.
|
| 316 |
-
"loss": 0.
|
| 317 |
"step": 44
|
| 318 |
},
|
| 319 |
{
|
| 320 |
-
"epoch": 1.
|
| 321 |
-
"grad_norm": 0.
|
| 322 |
-
"learning_rate": 7.
|
| 323 |
-
"loss": 0.
|
| 324 |
"step": 45
|
| 325 |
},
|
| 326 |
{
|
| 327 |
-
"epoch": 1.
|
| 328 |
-
"grad_norm": 0.
|
| 329 |
-
"learning_rate":
|
| 330 |
-
"loss": 0.
|
| 331 |
"step": 46
|
| 332 |
},
|
| 333 |
{
|
| 334 |
-
"epoch": 1.
|
| 335 |
-
"grad_norm": 0.
|
| 336 |
-
"learning_rate": 6.
|
| 337 |
-
"loss": 0.
|
| 338 |
"step": 47
|
| 339 |
},
|
| 340 |
{
|
| 341 |
-
"epoch": 1.
|
| 342 |
-
"grad_norm": 0.
|
| 343 |
-
"learning_rate": 6.
|
| 344 |
-
"loss": 0.
|
| 345 |
"step": 48
|
| 346 |
},
|
| 347 |
{
|
| 348 |
-
"epoch": 1.
|
| 349 |
-
"grad_norm": 0.
|
| 350 |
-
"learning_rate": 6.
|
| 351 |
-
"loss": 0.
|
| 352 |
"step": 49
|
| 353 |
},
|
| 354 |
{
|
| 355 |
-
"epoch": 1.
|
| 356 |
-
"grad_norm": 0.
|
| 357 |
-
"learning_rate": 6.
|
| 358 |
-
"loss": 0.
|
| 359 |
"step": 50
|
| 360 |
},
|
| 361 |
{
|
| 362 |
-
"epoch": 1.
|
| 363 |
-
"grad_norm": 0.
|
| 364 |
-
"learning_rate": 6.
|
| 365 |
-
"loss": 0.
|
| 366 |
"step": 51
|
| 367 |
},
|
| 368 |
{
|
| 369 |
-
"epoch": 1.
|
| 370 |
-
"grad_norm": 0.
|
| 371 |
-
"learning_rate":
|
| 372 |
-
"loss": 0.
|
| 373 |
"step": 52
|
| 374 |
},
|
| 375 |
{
|
| 376 |
-
"epoch": 1.
|
| 377 |
-
"grad_norm": 0.
|
| 378 |
-
"learning_rate":
|
| 379 |
-
"loss": 0.
|
| 380 |
"step": 53
|
| 381 |
},
|
| 382 |
{
|
| 383 |
-
"epoch": 1.
|
| 384 |
-
"grad_norm": 0.
|
| 385 |
-
"learning_rate": 5.
|
| 386 |
-
"loss": 0.
|
| 387 |
"step": 54
|
| 388 |
},
|
| 389 |
{
|
| 390 |
-
"epoch": 1.
|
| 391 |
-
"grad_norm": 0.
|
| 392 |
-
"learning_rate": 5.
|
| 393 |
-
"loss": 0.
|
| 394 |
"step": 55
|
| 395 |
},
|
| 396 |
{
|
| 397 |
-
"epoch": 1.
|
| 398 |
-
"grad_norm": 0.
|
| 399 |
-
"learning_rate": 5.
|
| 400 |
-
"loss": 0.
|
| 401 |
"step": 56
|
| 402 |
},
|
| 403 |
{
|
| 404 |
-
"epoch": 1.
|
| 405 |
-
"grad_norm": 0.
|
| 406 |
-
"learning_rate": 5.
|
| 407 |
-
"loss": 0.
|
| 408 |
"step": 57
|
| 409 |
},
|
| 410 |
{
|
| 411 |
-
"epoch": 1.
|
| 412 |
-
"grad_norm": 0.
|
| 413 |
-
"learning_rate":
|
| 414 |
-
"loss": 0.
|
| 415 |
"step": 58
|
| 416 |
},
|
| 417 |
{
|
| 418 |
-
"epoch": 1.
|
| 419 |
-
"grad_norm": 0.
|
| 420 |
-
"learning_rate":
|
| 421 |
-
"loss": 0.
|
| 422 |
"step": 59
|
| 423 |
},
|
| 424 |
{
|
| 425 |
-
"epoch": 1.
|
| 426 |
-
"grad_norm": 0.
|
| 427 |
-
"learning_rate": 4.
|
| 428 |
-
"loss": 0.
|
| 429 |
"step": 60
|
| 430 |
},
|
| 431 |
{
|
| 432 |
-
"epoch": 1.
|
| 433 |
-
"grad_norm": 0.
|
| 434 |
-
"learning_rate": 4.
|
| 435 |
-
"loss": 0.
|
| 436 |
"step": 61
|
| 437 |
},
|
| 438 |
{
|
| 439 |
-
"epoch": 1.
|
| 440 |
-
"grad_norm": 0.
|
| 441 |
-
"learning_rate": 4.
|
| 442 |
-
"loss": 0.
|
| 443 |
"step": 62
|
| 444 |
},
|
| 445 |
{
|
| 446 |
-
"epoch": 1.
|
| 447 |
-
"grad_norm": 0.
|
| 448 |
-
"learning_rate": 4.
|
| 449 |
-
"loss": 0.
|
| 450 |
"step": 63
|
| 451 |
},
|
| 452 |
{
|
| 453 |
-
"epoch": 1.
|
| 454 |
-
"grad_norm": 0.
|
| 455 |
-
"learning_rate": 4.
|
| 456 |
-
"loss": 0.
|
| 457 |
"step": 64
|
| 458 |
},
|
| 459 |
{
|
| 460 |
-
"epoch": 1.
|
| 461 |
-
"grad_norm": 0.
|
| 462 |
-
"learning_rate":
|
| 463 |
-
"loss": 0.
|
| 464 |
"step": 65
|
| 465 |
},
|
| 466 |
{
|
| 467 |
-
"epoch": 1.
|
| 468 |
-
"grad_norm": 0.
|
| 469 |
-
"learning_rate": 3.
|
| 470 |
-
"loss": 0.
|
| 471 |
"step": 66
|
| 472 |
},
|
| 473 |
{
|
| 474 |
-
"epoch": 1.
|
| 475 |
-
"grad_norm": 0.
|
| 476 |
-
"learning_rate": 3.
|
| 477 |
-
"loss": 0.
|
| 478 |
"step": 67
|
| 479 |
},
|
| 480 |
{
|
| 481 |
-
"epoch": 1.
|
| 482 |
-
"grad_norm": 0.
|
| 483 |
-
"learning_rate": 3.
|
| 484 |
-
"loss": 0.
|
| 485 |
"step": 68
|
| 486 |
},
|
| 487 |
{
|
| 488 |
-
"epoch": 1.
|
| 489 |
-
"grad_norm": 0.
|
| 490 |
-
"learning_rate": 3.
|
| 491 |
-
"loss": 0.
|
| 492 |
"step": 69
|
| 493 |
},
|
| 494 |
{
|
| 495 |
-
"epoch": 1.
|
| 496 |
-
"grad_norm": 0.
|
| 497 |
-
"learning_rate": 3.
|
| 498 |
-
"loss": 0.
|
| 499 |
"step": 70
|
| 500 |
},
|
| 501 |
{
|
| 502 |
-
"epoch":
|
| 503 |
-
"grad_norm":
|
| 504 |
-
"learning_rate":
|
| 505 |
-
"loss": 0.
|
| 506 |
"step": 71
|
| 507 |
},
|
| 508 |
{
|
| 509 |
-
"epoch": 2.
|
| 510 |
-
"grad_norm": 0.
|
| 511 |
-
"learning_rate":
|
| 512 |
-
"loss": 0.
|
| 513 |
"step": 72
|
| 514 |
},
|
| 515 |
{
|
| 516 |
-
"epoch": 2.
|
| 517 |
-
"grad_norm": 0.
|
| 518 |
-
"learning_rate": 2.
|
| 519 |
-
"loss": 0.
|
| 520 |
"step": 73
|
| 521 |
},
|
| 522 |
{
|
| 523 |
-
"epoch": 2.
|
| 524 |
-
"grad_norm": 0.
|
| 525 |
-
"learning_rate": 2.
|
| 526 |
-
"loss": 0.
|
| 527 |
"step": 74
|
| 528 |
},
|
| 529 |
{
|
| 530 |
-
"epoch": 2.
|
| 531 |
-
"grad_norm": 0.
|
| 532 |
-
"learning_rate": 2.
|
| 533 |
-
"loss": 0.
|
| 534 |
"step": 75
|
| 535 |
},
|
| 536 |
{
|
| 537 |
-
"epoch": 2.
|
| 538 |
-
"grad_norm": 0.
|
| 539 |
-
"learning_rate": 2.
|
| 540 |
-
"loss": 0.
|
| 541 |
"step": 76
|
| 542 |
},
|
| 543 |
{
|
| 544 |
-
"epoch": 2.
|
| 545 |
-
"grad_norm": 0.
|
| 546 |
-
"learning_rate": 2.
|
| 547 |
-
"loss": 0.
|
| 548 |
"step": 77
|
| 549 |
},
|
| 550 |
{
|
| 551 |
-
"epoch": 2.
|
| 552 |
-
"grad_norm":
|
| 553 |
-
"learning_rate":
|
| 554 |
-
"loss": 0.
|
| 555 |
"step": 78
|
| 556 |
},
|
| 557 |
{
|
| 558 |
-
"epoch": 2.
|
| 559 |
-
"grad_norm": 0.
|
| 560 |
-
"learning_rate":
|
| 561 |
-
"loss": 0.
|
| 562 |
"step": 79
|
| 563 |
},
|
| 564 |
{
|
| 565 |
-
"epoch": 2.
|
| 566 |
-
"grad_norm":
|
| 567 |
-
"learning_rate": 1.
|
| 568 |
-
"loss": 0.
|
| 569 |
"step": 80
|
| 570 |
},
|
| 571 |
{
|
| 572 |
-
"epoch": 2.
|
| 573 |
-
"grad_norm": 0.
|
| 574 |
-
"learning_rate": 1.
|
| 575 |
-
"loss": 0.
|
| 576 |
"step": 81
|
| 577 |
},
|
| 578 |
{
|
| 579 |
-
"epoch": 2.
|
| 580 |
-
"grad_norm": 0.
|
| 581 |
-
"learning_rate": 1.
|
| 582 |
-
"loss": 0.
|
| 583 |
"step": 82
|
| 584 |
},
|
| 585 |
{
|
| 586 |
-
"epoch": 2.
|
| 587 |
-
"grad_norm": 0.
|
| 588 |
-
"learning_rate": 1.
|
| 589 |
-
"loss": 0.
|
| 590 |
"step": 83
|
| 591 |
},
|
| 592 |
{
|
| 593 |
-
"epoch": 2.
|
| 594 |
-
"grad_norm": 0.
|
| 595 |
-
"learning_rate": 1.
|
| 596 |
-
"loss": 0.
|
| 597 |
"step": 84
|
| 598 |
},
|
| 599 |
{
|
| 600 |
-
"epoch": 2.
|
| 601 |
-
"grad_norm": 0.
|
| 602 |
-
"learning_rate": 1.
|
| 603 |
-
"loss": 0.
|
| 604 |
"step": 85
|
| 605 |
},
|
| 606 |
{
|
| 607 |
-
"epoch": 2.
|
| 608 |
-
"grad_norm": 0.
|
| 609 |
-
"learning_rate":
|
| 610 |
-
"loss": 0.
|
| 611 |
"step": 86
|
| 612 |
},
|
| 613 |
{
|
| 614 |
-
"epoch": 2.
|
| 615 |
-
"grad_norm": 0.
|
| 616 |
-
"learning_rate":
|
| 617 |
-
"loss": 0.
|
| 618 |
"step": 87
|
| 619 |
},
|
| 620 |
{
|
| 621 |
-
"epoch": 2.
|
| 622 |
-
"grad_norm": 0.
|
| 623 |
-
"learning_rate":
|
| 624 |
-
"loss": 0.
|
| 625 |
"step": 88
|
| 626 |
},
|
| 627 |
{
|
| 628 |
-
"epoch": 2.
|
| 629 |
-
"grad_norm": 0.
|
| 630 |
-
"learning_rate":
|
| 631 |
-
"loss": 0.
|
| 632 |
"step": 89
|
| 633 |
},
|
| 634 |
{
|
| 635 |
-
"epoch": 2.
|
| 636 |
-
"grad_norm": 0.
|
| 637 |
-
"learning_rate":
|
| 638 |
-
"loss": 0.
|
| 639 |
"step": 90
|
| 640 |
},
|
| 641 |
{
|
| 642 |
-
"epoch": 2.
|
| 643 |
-
"grad_norm": 0.
|
| 644 |
-
"learning_rate":
|
| 645 |
-
"loss": 0.
|
| 646 |
"step": 91
|
| 647 |
},
|
| 648 |
{
|
| 649 |
-
"epoch": 2.
|
| 650 |
-
"grad_norm": 0.
|
| 651 |
-
"learning_rate":
|
| 652 |
-
"loss": 0.
|
| 653 |
"step": 92
|
| 654 |
},
|
| 655 |
{
|
| 656 |
-
"epoch": 2.
|
| 657 |
-
"grad_norm": 0.
|
| 658 |
-
"learning_rate":
|
| 659 |
-
"loss": 0.
|
| 660 |
"step": 93
|
| 661 |
},
|
| 662 |
{
|
| 663 |
-
"epoch": 2.
|
| 664 |
-
"grad_norm": 0.
|
| 665 |
-
"learning_rate":
|
| 666 |
-
"loss": 0.
|
| 667 |
"step": 94
|
| 668 |
},
|
| 669 |
{
|
| 670 |
-
"epoch": 2.
|
| 671 |
-
"grad_norm": 0.
|
| 672 |
-
"learning_rate":
|
| 673 |
-
"loss": 0.
|
| 674 |
"step": 95
|
| 675 |
},
|
| 676 |
{
|
| 677 |
-
"epoch": 2.
|
| 678 |
-
"grad_norm": 0.
|
| 679 |
-
"learning_rate":
|
| 680 |
-
"loss": 0.
|
| 681 |
"step": 96
|
| 682 |
},
|
| 683 |
{
|
| 684 |
-
"epoch": 2.
|
| 685 |
-
"grad_norm": 0.
|
| 686 |
-
"learning_rate":
|
| 687 |
-
"loss": 0.
|
| 688 |
"step": 97
|
| 689 |
},
|
| 690 |
{
|
| 691 |
-
"epoch": 2.
|
| 692 |
-
"grad_norm": 0.
|
| 693 |
-
"learning_rate":
|
| 694 |
-
"loss": 0.
|
| 695 |
"step": 98
|
| 696 |
},
|
| 697 |
{
|
| 698 |
-
"epoch": 2.
|
| 699 |
-
"grad_norm": 0.
|
| 700 |
-
"learning_rate":
|
| 701 |
-
"loss": 0.
|
| 702 |
"step": 99
|
| 703 |
},
|
| 704 |
{
|
| 705 |
-
"epoch": 2.
|
| 706 |
-
"grad_norm": 0.
|
| 707 |
-
"learning_rate":
|
| 708 |
-
"loss": 0.
|
| 709 |
"step": 100
|
| 710 |
},
|
| 711 |
{
|
| 712 |
-
"epoch": 2.
|
| 713 |
-
"grad_norm": 0.
|
| 714 |
-
"learning_rate":
|
| 715 |
-
"loss": 0.
|
| 716 |
"step": 101
|
| 717 |
},
|
| 718 |
{
|
| 719 |
-
"epoch": 2.
|
| 720 |
-
"grad_norm": 0.
|
| 721 |
-
"learning_rate":
|
| 722 |
-
"loss": 0.
|
| 723 |
"step": 102
|
| 724 |
},
|
| 725 |
{
|
| 726 |
-
"epoch": 2.
|
| 727 |
-
"grad_norm": 0.
|
| 728 |
-
"learning_rate":
|
| 729 |
-
"loss": 0.
|
| 730 |
"step": 103
|
| 731 |
},
|
| 732 |
{
|
| 733 |
-
"epoch": 2.
|
| 734 |
-
"grad_norm": 0.
|
| 735 |
-
"learning_rate":
|
| 736 |
-
"loss": 0.
|
| 737 |
"step": 104
|
| 738 |
},
|
| 739 |
{
|
| 740 |
-
"epoch": 2.
|
| 741 |
-
"grad_norm": 0.
|
| 742 |
-
"learning_rate": 2.358289416693027e-08,
|
| 743 |
-
"loss": 0.4073,
|
| 744 |
-
"step": 105
|
| 745 |
-
},
|
| 746 |
-
{
|
| 747 |
-
"epoch": 2.9444444444444446,
|
| 748 |
-
"grad_norm": 0.6272555116057572,
|
| 749 |
-
"learning_rate": 1.0485868811441757e-08,
|
| 750 |
-
"loss": 0.4084,
|
| 751 |
-
"step": 106
|
| 752 |
-
},
|
| 753 |
-
{
|
| 754 |
-
"epoch": 2.9722222222222223,
|
| 755 |
-
"grad_norm": 0.6263081374423879,
|
| 756 |
-
"learning_rate": 2.6221547724253337e-09,
|
| 757 |
-
"loss": 0.3967,
|
| 758 |
-
"step": 107
|
| 759 |
-
},
|
| 760 |
-
{
|
| 761 |
-
"epoch": 3.0,
|
| 762 |
-
"grad_norm": 0.5852999921009103,
|
| 763 |
"learning_rate": 0.0,
|
| 764 |
-
"loss": 0.
|
| 765 |
-
"step":
|
| 766 |
},
|
| 767 |
{
|
| 768 |
-
"epoch":
|
| 769 |
-
"step":
|
| 770 |
-
"total_flos":
|
| 771 |
-
"train_loss": 0.
|
| 772 |
-
"train_runtime":
|
| 773 |
-
"train_samples_per_second":
|
| 774 |
-
"train_steps_per_second": 0.
|
| 775 |
}
|
| 776 |
],
|
| 777 |
"logging_steps": 1,
|
| 778 |
-
"max_steps":
|
| 779 |
"num_input_tokens_seen": 0,
|
| 780 |
"num_train_epochs": 3,
|
| 781 |
"save_steps": 500,
|
|
@@ -791,8 +770,8 @@
|
|
| 791 |
"attributes": {}
|
| 792 |
}
|
| 793 |
},
|
| 794 |
-
"total_flos":
|
| 795 |
-
"train_batch_size":
|
| 796 |
"trial_name": null,
|
| 797 |
"trial_params": null
|
| 798 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"best_metric": null,
|
| 3 |
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 2.9626168224299065,
|
| 5 |
"eval_steps": 500,
|
| 6 |
+
"global_step": 105,
|
| 7 |
"is_hyper_param_search": false,
|
| 8 |
"is_local_process_zero": true,
|
| 9 |
"is_world_process_zero": true,
|
| 10 |
"log_history": [
|
| 11 |
{
|
| 12 |
+
"epoch": 0.028037383177570093,
|
| 13 |
+
"grad_norm": 5.46907120535221,
|
| 14 |
"learning_rate": 9.090909090909091e-07,
|
| 15 |
+
"loss": 0.6948,
|
| 16 |
"step": 1
|
| 17 |
},
|
| 18 |
{
|
| 19 |
+
"epoch": 0.056074766355140186,
|
| 20 |
+
"grad_norm": 5.695701169228273,
|
| 21 |
"learning_rate": 1.8181818181818183e-06,
|
| 22 |
+
"loss": 0.659,
|
| 23 |
"step": 2
|
| 24 |
},
|
| 25 |
{
|
| 26 |
+
"epoch": 0.08411214953271028,
|
| 27 |
+
"grad_norm": 5.869868275090631,
|
| 28 |
"learning_rate": 2.7272727272727272e-06,
|
| 29 |
+
"loss": 0.6481,
|
| 30 |
"step": 3
|
| 31 |
},
|
| 32 |
{
|
| 33 |
+
"epoch": 0.11214953271028037,
|
| 34 |
+
"grad_norm": 5.197705973950515,
|
| 35 |
"learning_rate": 3.6363636363636366e-06,
|
| 36 |
+
"loss": 0.6641,
|
| 37 |
"step": 4
|
| 38 |
},
|
| 39 |
{
|
| 40 |
+
"epoch": 0.14018691588785046,
|
| 41 |
+
"grad_norm": 3.3963665075709906,
|
| 42 |
"learning_rate": 4.5454545454545455e-06,
|
| 43 |
+
"loss": 0.5579,
|
| 44 |
"step": 5
|
| 45 |
},
|
| 46 |
{
|
| 47 |
+
"epoch": 0.16822429906542055,
|
| 48 |
+
"grad_norm": 2.20924738877731,
|
| 49 |
"learning_rate": 5.4545454545454545e-06,
|
| 50 |
+
"loss": 0.6036,
|
| 51 |
"step": 6
|
| 52 |
},
|
| 53 |
{
|
| 54 |
+
"epoch": 0.19626168224299065,
|
| 55 |
+
"grad_norm": 1.984739518286341,
|
| 56 |
"learning_rate": 6.363636363636364e-06,
|
| 57 |
+
"loss": 0.5288,
|
| 58 |
"step": 7
|
| 59 |
},
|
| 60 |
{
|
| 61 |
+
"epoch": 0.22429906542056074,
|
| 62 |
+
"grad_norm": 1.9291295994072917,
|
| 63 |
"learning_rate": 7.272727272727273e-06,
|
| 64 |
"loss": 0.5054,
|
| 65 |
"step": 8
|
| 66 |
},
|
| 67 |
{
|
| 68 |
+
"epoch": 0.2523364485981308,
|
| 69 |
+
"grad_norm": 1.6389948204701459,
|
| 70 |
"learning_rate": 8.181818181818183e-06,
|
| 71 |
+
"loss": 0.5339,
|
| 72 |
"step": 9
|
| 73 |
},
|
| 74 |
{
|
| 75 |
+
"epoch": 0.2803738317757009,
|
| 76 |
+
"grad_norm": 1.4300152832592459,
|
| 77 |
"learning_rate": 9.090909090909091e-06,
|
| 78 |
"loss": 0.5433,
|
| 79 |
"step": 10
|
| 80 |
},
|
| 81 |
{
|
| 82 |
+
"epoch": 0.308411214953271,
|
| 83 |
+
"grad_norm": 1.711886946596263,
|
| 84 |
"learning_rate": 1e-05,
|
| 85 |
+
"loss": 0.5537,
|
| 86 |
"step": 11
|
| 87 |
},
|
| 88 |
{
|
| 89 |
+
"epoch": 0.3364485981308411,
|
| 90 |
+
"grad_norm": 1.5442931844274987,
|
| 91 |
+
"learning_rate": 9.997207818651273e-06,
|
| 92 |
+
"loss": 0.498,
|
| 93 |
"step": 12
|
| 94 |
},
|
| 95 |
{
|
| 96 |
+
"epoch": 0.3644859813084112,
|
| 97 |
+
"grad_norm": 1.3068883684134325,
|
| 98 |
+
"learning_rate": 9.988834393115768e-06,
|
| 99 |
+
"loss": 0.6412,
|
| 100 |
"step": 13
|
| 101 |
},
|
| 102 |
{
|
| 103 |
+
"epoch": 0.3925233644859813,
|
| 104 |
+
"grad_norm": 1.2741183296138552,
|
| 105 |
+
"learning_rate": 9.97488907544252e-06,
|
| 106 |
+
"loss": 0.5796,
|
| 107 |
"step": 14
|
| 108 |
},
|
| 109 |
{
|
| 110 |
+
"epoch": 0.4205607476635514,
|
| 111 |
+
"grad_norm": 1.0164680008182152,
|
| 112 |
+
"learning_rate": 9.955387440773902e-06,
|
| 113 |
+
"loss": 0.5019,
|
| 114 |
"step": 15
|
| 115 |
},
|
| 116 |
{
|
| 117 |
+
"epoch": 0.4485981308411215,
|
| 118 |
+
"grad_norm": 1.036540358883728,
|
| 119 |
+
"learning_rate": 9.930351269950144e-06,
|
| 120 |
+
"loss": 0.5982,
|
| 121 |
"step": 16
|
| 122 |
},
|
| 123 |
{
|
| 124 |
+
"epoch": 0.4766355140186916,
|
| 125 |
+
"grad_norm": 1.0480857466954097,
|
| 126 |
+
"learning_rate": 9.899808525182935e-06,
|
| 127 |
+
"loss": 0.5377,
|
| 128 |
"step": 17
|
| 129 |
},
|
| 130 |
{
|
| 131 |
+
"epoch": 0.5046728971962616,
|
| 132 |
+
"grad_norm": 0.8148023572178399,
|
| 133 |
+
"learning_rate": 9.863793318825186e-06,
|
| 134 |
+
"loss": 0.4776,
|
| 135 |
"step": 18
|
| 136 |
},
|
| 137 |
{
|
| 138 |
+
"epoch": 0.5327102803738317,
|
| 139 |
+
"grad_norm": 1.0017091343990974,
|
| 140 |
+
"learning_rate": 9.822345875271884e-06,
|
| 141 |
+
"loss": 0.648,
|
| 142 |
"step": 19
|
| 143 |
},
|
| 144 |
{
|
| 145 |
+
"epoch": 0.5607476635514018,
|
| 146 |
+
"grad_norm": 0.7723478036715665,
|
| 147 |
+
"learning_rate": 9.775512486034564e-06,
|
| 148 |
+
"loss": 0.4544,
|
| 149 |
"step": 20
|
| 150 |
},
|
| 151 |
{
|
| 152 |
+
"epoch": 0.5887850467289719,
|
| 153 |
+
"grad_norm": 0.8710833273679682,
|
| 154 |
+
"learning_rate": 9.723345458039595e-06,
|
| 155 |
+
"loss": 0.4936,
|
| 156 |
"step": 21
|
| 157 |
},
|
| 158 |
{
|
| 159 |
+
"epoch": 0.616822429906542,
|
| 160 |
+
"grad_norm": 0.9055028217774582,
|
| 161 |
+
"learning_rate": 9.665903055208013e-06,
|
| 162 |
+
"loss": 0.5819,
|
| 163 |
"step": 22
|
| 164 |
},
|
| 165 |
{
|
| 166 |
+
"epoch": 0.6448598130841121,
|
| 167 |
+
"grad_norm": 0.8103202901623571,
|
| 168 |
+
"learning_rate": 9.603249433382145e-06,
|
| 169 |
+
"loss": 0.5506,
|
| 170 |
"step": 23
|
| 171 |
},
|
| 172 |
{
|
| 173 |
+
"epoch": 0.6728971962616822,
|
| 174 |
+
"grad_norm": 0.7712121760995281,
|
| 175 |
+
"learning_rate": 9.535454568671705e-06,
|
| 176 |
+
"loss": 0.4932,
|
| 177 |
"step": 24
|
| 178 |
},
|
| 179 |
{
|
| 180 |
+
"epoch": 0.7009345794392523,
|
| 181 |
+
"grad_norm": 0.9091205022340342,
|
| 182 |
+
"learning_rate": 9.462594179299408e-06,
|
| 183 |
+
"loss": 0.5776,
|
| 184 |
"step": 25
|
| 185 |
},
|
| 186 |
{
|
| 187 |
+
"epoch": 0.7289719626168224,
|
| 188 |
+
"grad_norm": 0.7512412627292558,
|
| 189 |
+
"learning_rate": 9.384749641033358e-06,
|
| 190 |
+
"loss": 0.4665,
|
| 191 |
"step": 26
|
| 192 |
},
|
| 193 |
{
|
| 194 |
+
"epoch": 0.7570093457943925,
|
| 195 |
+
"grad_norm": 0.9759396921985853,
|
| 196 |
+
"learning_rate": 9.302007896300697e-06,
|
| 197 |
+
"loss": 0.6547,
|
| 198 |
"step": 27
|
| 199 |
},
|
| 200 |
{
|
| 201 |
+
"epoch": 0.7850467289719626,
|
| 202 |
+
"grad_norm": 0.8218161831230707,
|
| 203 |
+
"learning_rate": 9.214461357083986e-06,
|
| 204 |
+
"loss": 0.5403,
|
| 205 |
"step": 28
|
| 206 |
},
|
| 207 |
{
|
| 208 |
+
"epoch": 0.8130841121495327,
|
| 209 |
+
"grad_norm": 0.7802775736672756,
|
| 210 |
+
"learning_rate": 9.122207801708802e-06,
|
| 211 |
+
"loss": 0.4792,
|
| 212 |
"step": 29
|
| 213 |
},
|
| 214 |
{
|
| 215 |
+
"epoch": 0.8411214953271028,
|
| 216 |
+
"grad_norm": 1.2211854499314727,
|
| 217 |
+
"learning_rate": 9.025350265637816e-06,
|
| 218 |
+
"loss": 0.5767,
|
| 219 |
"step": 30
|
| 220 |
},
|
| 221 |
{
|
| 222 |
+
"epoch": 0.8691588785046729,
|
| 223 |
+
"grad_norm": 0.7291315847510311,
|
| 224 |
+
"learning_rate": 8.923996926393306e-06,
|
| 225 |
+
"loss": 0.5339,
|
| 226 |
"step": 31
|
| 227 |
},
|
| 228 |
{
|
| 229 |
+
"epoch": 0.897196261682243,
|
| 230 |
+
"grad_norm": 0.7685026874314949,
|
| 231 |
+
"learning_rate": 8.818260982736662e-06,
|
| 232 |
+
"loss": 0.5102,
|
| 233 |
"step": 32
|
| 234 |
},
|
| 235 |
{
|
| 236 |
+
"epoch": 0.9252336448598131,
|
| 237 |
+
"grad_norm": 0.7819112577141099,
|
| 238 |
+
"learning_rate": 8.708260528239788e-06,
|
| 239 |
+
"loss": 0.5196,
|
| 240 |
"step": 33
|
| 241 |
},
|
| 242 |
{
|
| 243 |
+
"epoch": 0.9532710280373832,
|
| 244 |
+
"grad_norm": 0.805804762659976,
|
| 245 |
+
"learning_rate": 8.594118419389648e-06,
|
| 246 |
+
"loss": 0.5486,
|
| 247 |
"step": 34
|
| 248 |
},
|
| 249 |
{
|
| 250 |
+
"epoch": 0.9813084112149533,
|
| 251 |
+
"grad_norm": 0.6960763295513189,
|
| 252 |
+
"learning_rate": 8.475962138373212e-06,
|
| 253 |
+
"loss": 0.4885,
|
| 254 |
"step": 35
|
| 255 |
},
|
| 256 |
{
|
| 257 |
+
"epoch": 1.0186915887850467,
|
| 258 |
+
"grad_norm": 1.261779486223386,
|
| 259 |
+
"learning_rate": 8.353923650696119e-06,
|
| 260 |
+
"loss": 0.7283,
|
| 261 |
"step": 36
|
| 262 |
},
|
| 263 |
{
|
| 264 |
+
"epoch": 1.0467289719626167,
|
| 265 |
+
"grad_norm": 2.1904861480940703,
|
| 266 |
+
"learning_rate": 8.228139257794012e-06,
|
| 267 |
+
"loss": 0.5858,
|
| 268 |
"step": 37
|
| 269 |
},
|
| 270 |
{
|
| 271 |
+
"epoch": 1.074766355140187,
|
| 272 |
+
"grad_norm": 0.836565359732725,
|
| 273 |
+
"learning_rate": 8.098749444801226e-06,
|
| 274 |
+
"loss": 0.4984,
|
| 275 |
"step": 38
|
| 276 |
},
|
| 277 |
{
|
| 278 |
+
"epoch": 1.102803738317757,
|
| 279 |
+
"grad_norm": 0.7156392071456283,
|
| 280 |
+
"learning_rate": 7.965898723646777e-06,
|
| 281 |
+
"loss": 0.4133,
|
| 282 |
"step": 39
|
| 283 |
},
|
| 284 |
{
|
| 285 |
+
"epoch": 1.1308411214953271,
|
| 286 |
+
"grad_norm": 0.7811150734148334,
|
| 287 |
+
"learning_rate": 7.829735471652978e-06,
|
| 288 |
+
"loss": 0.4737,
|
| 289 |
"step": 40
|
| 290 |
},
|
| 291 |
{
|
| 292 |
+
"epoch": 1.158878504672897,
|
| 293 |
+
"grad_norm": 0.8005430413673945,
|
| 294 |
+
"learning_rate": 7.690411765816864e-06,
|
| 295 |
+
"loss": 0.461,
|
| 296 |
"step": 41
|
| 297 |
},
|
| 298 |
{
|
| 299 |
+
"epoch": 1.1869158878504673,
|
| 300 |
+
"grad_norm": 0.734823009686673,
|
| 301 |
+
"learning_rate": 7.548083212959588e-06,
|
| 302 |
+
"loss": 0.4878,
|
| 303 |
"step": 42
|
| 304 |
},
|
| 305 |
{
|
| 306 |
+
"epoch": 1.2149532710280373,
|
| 307 |
+
"grad_norm": 0.9106888643783978,
|
| 308 |
+
"learning_rate": 7.402908775933419e-06,
|
| 309 |
+
"loss": 0.5398,
|
| 310 |
"step": 43
|
| 311 |
},
|
| 312 |
{
|
| 313 |
+
"epoch": 1.2429906542056075,
|
| 314 |
+
"grad_norm": 0.6471585113802202,
|
| 315 |
+
"learning_rate": 7.25505059608051e-06,
|
| 316 |
+
"loss": 0.4282,
|
| 317 |
"step": 44
|
| 318 |
},
|
| 319 |
{
|
| 320 |
+
"epoch": 1.2710280373831775,
|
| 321 |
+
"grad_norm": 0.7710047485521259,
|
| 322 |
+
"learning_rate": 7.104673812141676e-06,
|
| 323 |
+
"loss": 0.4352,
|
| 324 |
"step": 45
|
| 325 |
},
|
| 326 |
{
|
| 327 |
+
"epoch": 1.2990654205607477,
|
| 328 |
+
"grad_norm": 0.6543038711142444,
|
| 329 |
+
"learning_rate": 6.9519463758174745e-06,
|
| 330 |
+
"loss": 0.3746,
|
| 331 |
"step": 46
|
| 332 |
},
|
| 333 |
{
|
| 334 |
+
"epoch": 1.3271028037383177,
|
| 335 |
+
"grad_norm": 0.6480235385313834,
|
| 336 |
+
"learning_rate": 6.797038864187564e-06,
|
| 337 |
+
"loss": 0.4648,
|
| 338 |
"step": 47
|
| 339 |
},
|
| 340 |
{
|
| 341 |
+
"epoch": 1.355140186915888,
|
| 342 |
+
"grad_norm": 0.663332249258245,
|
| 343 |
+
"learning_rate": 6.640124289197845e-06,
|
| 344 |
+
"loss": 0.412,
|
| 345 |
"step": 48
|
| 346 |
},
|
| 347 |
{
|
| 348 |
+
"epoch": 1.3831775700934579,
|
| 349 |
+
"grad_norm": 0.735147105832247,
|
| 350 |
+
"learning_rate": 6.481377904428171e-06,
|
| 351 |
+
"loss": 0.4639,
|
| 352 |
"step": 49
|
| 353 |
},
|
| 354 |
{
|
| 355 |
+
"epoch": 1.411214953271028,
|
| 356 |
+
"grad_norm": 0.7169861474373482,
|
| 357 |
+
"learning_rate": 6.3209770093564315e-06,
|
| 358 |
+
"loss": 0.4511,
|
| 359 |
"step": 50
|
| 360 |
},
|
| 361 |
{
|
| 362 |
+
"epoch": 1.439252336448598,
|
| 363 |
+
"grad_norm": 0.7241007166406079,
|
| 364 |
+
"learning_rate": 6.1591007513376425e-06,
|
| 365 |
+
"loss": 0.4204,
|
| 366 |
"step": 51
|
| 367 |
},
|
| 368 |
{
|
| 369 |
+
"epoch": 1.4672897196261683,
|
| 370 |
+
"grad_norm": 0.7785664265979848,
|
| 371 |
+
"learning_rate": 5.995929925519181e-06,
|
| 372 |
+
"loss": 0.4717,
|
| 373 |
"step": 52
|
| 374 |
},
|
| 375 |
{
|
| 376 |
+
"epoch": 1.4953271028037383,
|
| 377 |
+
"grad_norm": 0.643031905575638,
|
| 378 |
+
"learning_rate": 5.831646772915651e-06,
|
| 379 |
+
"loss": 0.451,
|
| 380 |
"step": 53
|
| 381 |
},
|
| 382 |
{
|
| 383 |
+
"epoch": 1.5233644859813085,
|
| 384 |
+
"grad_norm": 0.6622611979477618,
|
| 385 |
+
"learning_rate": 5.666434776868895e-06,
|
| 386 |
+
"loss": 0.4598,
|
| 387 |
"step": 54
|
| 388 |
},
|
| 389 |
{
|
| 390 |
+
"epoch": 1.5514018691588785,
|
| 391 |
+
"grad_norm": 0.6681223638603929,
|
| 392 |
+
"learning_rate": 5.500478458120493e-06,
|
| 393 |
+
"loss": 0.4471,
|
| 394 |
"step": 55
|
| 395 |
},
|
| 396 |
{
|
| 397 |
+
"epoch": 1.5794392523364484,
|
| 398 |
+
"grad_norm": 0.6742446596507364,
|
| 399 |
+
"learning_rate": 5.3339631687256085e-06,
|
| 400 |
+
"loss": 0.4323,
|
| 401 |
"step": 56
|
| 402 |
},
|
| 403 |
{
|
| 404 |
+
"epoch": 1.6074766355140186,
|
| 405 |
+
"grad_norm": 0.6011350150512597,
|
| 406 |
+
"learning_rate": 5.1670748850383734e-06,
|
| 407 |
+
"loss": 0.421,
|
| 408 |
"step": 57
|
| 409 |
},
|
| 410 |
{
|
| 411 |
+
"epoch": 1.6355140186915889,
|
| 412 |
+
"grad_norm": 0.6177912691346672,
|
| 413 |
+
"learning_rate": 5e-06,
|
| 414 |
+
"loss": 0.3956,
|
| 415 |
"step": 58
|
| 416 |
},
|
| 417 |
{
|
| 418 |
+
"epoch": 1.6635514018691588,
|
| 419 |
+
"grad_norm": 0.6868626354051661,
|
| 420 |
+
"learning_rate": 4.832925114961629e-06,
|
| 421 |
+
"loss": 0.4661,
|
| 422 |
"step": 59
|
| 423 |
},
|
| 424 |
{
|
| 425 |
+
"epoch": 1.6915887850467288,
|
| 426 |
+
"grad_norm": 0.6421560332745099,
|
| 427 |
+
"learning_rate": 4.666036831274392e-06,
|
| 428 |
+
"loss": 0.4704,
|
| 429 |
"step": 60
|
| 430 |
},
|
| 431 |
{
|
| 432 |
+
"epoch": 1.719626168224299,
|
| 433 |
+
"grad_norm": 0.7187797397726619,
|
| 434 |
+
"learning_rate": 4.499521541879508e-06,
|
| 435 |
+
"loss": 0.465,
|
| 436 |
"step": 61
|
| 437 |
},
|
| 438 |
{
|
| 439 |
+
"epoch": 1.7476635514018692,
|
| 440 |
+
"grad_norm": 0.5867907128348048,
|
| 441 |
+
"learning_rate": 4.333565223131107e-06,
|
| 442 |
+
"loss": 0.3283,
|
| 443 |
"step": 62
|
| 444 |
},
|
| 445 |
{
|
| 446 |
+
"epoch": 1.7757009345794392,
|
| 447 |
+
"grad_norm": 0.662616949214837,
|
| 448 |
+
"learning_rate": 4.1683532270843505e-06,
|
| 449 |
+
"loss": 0.4216,
|
| 450 |
"step": 63
|
| 451 |
},
|
| 452 |
{
|
| 453 |
+
"epoch": 1.8037383177570092,
|
| 454 |
+
"grad_norm": 0.6719507669829111,
|
| 455 |
+
"learning_rate": 4.004070074480821e-06,
|
| 456 |
+
"loss": 0.4443,
|
| 457 |
"step": 64
|
| 458 |
},
|
| 459 |
{
|
| 460 |
+
"epoch": 1.8317757009345794,
|
| 461 |
+
"grad_norm": 0.6464555389487582,
|
| 462 |
+
"learning_rate": 3.840899248662358e-06,
|
| 463 |
+
"loss": 0.4901,
|
| 464 |
"step": 65
|
| 465 |
},
|
| 466 |
{
|
| 467 |
+
"epoch": 1.8598130841121496,
|
| 468 |
+
"grad_norm": 0.642004590459993,
|
| 469 |
+
"learning_rate": 3.6790229906435706e-06,
|
| 470 |
+
"loss": 0.4687,
|
| 471 |
"step": 66
|
| 472 |
},
|
| 473 |
{
|
| 474 |
+
"epoch": 1.8878504672897196,
|
| 475 |
+
"grad_norm": 0.6811855386223414,
|
| 476 |
+
"learning_rate": 3.518622095571831e-06,
|
| 477 |
+
"loss": 0.5103,
|
| 478 |
"step": 67
|
| 479 |
},
|
| 480 |
{
|
| 481 |
+
"epoch": 1.9158878504672896,
|
| 482 |
+
"grad_norm": 0.6780639408715025,
|
| 483 |
+
"learning_rate": 3.3598757108021546e-06,
|
| 484 |
+
"loss": 0.4563,
|
| 485 |
"step": 68
|
| 486 |
},
|
| 487 |
{
|
| 488 |
+
"epoch": 1.9439252336448598,
|
| 489 |
+
"grad_norm": 0.6096518872897251,
|
| 490 |
+
"learning_rate": 3.202961135812437e-06,
|
| 491 |
+
"loss": 0.4273,
|
| 492 |
"step": 69
|
| 493 |
},
|
| 494 |
{
|
| 495 |
+
"epoch": 1.97196261682243,
|
| 496 |
+
"grad_norm": 0.6148567020386766,
|
| 497 |
+
"learning_rate": 3.0480536241825263e-06,
|
| 498 |
+
"loss": 0.3991,
|
| 499 |
"step": 70
|
| 500 |
},
|
| 501 |
{
|
| 502 |
+
"epoch": 2.0093457943925235,
|
| 503 |
+
"grad_norm": 1.1093908775727706,
|
| 504 |
+
"learning_rate": 2.8953261878583263e-06,
|
| 505 |
+
"loss": 0.6552,
|
| 506 |
"step": 71
|
| 507 |
},
|
| 508 |
{
|
| 509 |
+
"epoch": 2.0373831775700935,
|
| 510 |
+
"grad_norm": 0.5907034426928125,
|
| 511 |
+
"learning_rate": 2.74494940391949e-06,
|
| 512 |
+
"loss": 0.4244,
|
| 513 |
"step": 72
|
| 514 |
},
|
| 515 |
{
|
| 516 |
+
"epoch": 2.0654205607476634,
|
| 517 |
+
"grad_norm": 0.6237639437264948,
|
| 518 |
+
"learning_rate": 2.5970912240665815e-06,
|
| 519 |
+
"loss": 0.3433,
|
| 520 |
"step": 73
|
| 521 |
},
|
| 522 |
{
|
| 523 |
+
"epoch": 2.0934579439252334,
|
| 524 |
+
"grad_norm": 0.6456026273029044,
|
| 525 |
+
"learning_rate": 2.4519167870404126e-06,
|
| 526 |
+
"loss": 0.3633,
|
| 527 |
"step": 74
|
| 528 |
},
|
| 529 |
{
|
| 530 |
+
"epoch": 2.121495327102804,
|
| 531 |
+
"grad_norm": 0.6114145984670124,
|
| 532 |
+
"learning_rate": 2.309588234183137e-06,
|
| 533 |
+
"loss": 0.3607,
|
| 534 |
"step": 75
|
| 535 |
},
|
| 536 |
{
|
| 537 |
+
"epoch": 2.149532710280374,
|
| 538 |
+
"grad_norm": 0.5622408748138118,
|
| 539 |
+
"learning_rate": 2.1702645283470238e-06,
|
| 540 |
+
"loss": 0.3255,
|
| 541 |
"step": 76
|
| 542 |
},
|
| 543 |
{
|
| 544 |
+
"epoch": 2.177570093457944,
|
| 545 |
+
"grad_norm": 0.6289348884143411,
|
| 546 |
+
"learning_rate": 2.0341012763532243e-06,
|
| 547 |
+
"loss": 0.3917,
|
| 548 |
"step": 77
|
| 549 |
},
|
| 550 |
{
|
| 551 |
+
"epoch": 2.205607476635514,
|
| 552 |
+
"grad_norm": 2.11632857452312,
|
| 553 |
+
"learning_rate": 1.9012505551987764e-06,
|
| 554 |
+
"loss": 0.5284,
|
| 555 |
"step": 78
|
| 556 |
},
|
| 557 |
{
|
| 558 |
+
"epoch": 2.2336448598130842,
|
| 559 |
+
"grad_norm": 0.6011421789250062,
|
| 560 |
+
"learning_rate": 1.771860742205988e-06,
|
| 561 |
+
"loss": 0.3717,
|
| 562 |
"step": 79
|
| 563 |
},
|
| 564 |
{
|
| 565 |
+
"epoch": 2.2616822429906542,
|
| 566 |
+
"grad_norm": 0.605196953414431,
|
| 567 |
+
"learning_rate": 1.646076349303884e-06,
|
| 568 |
+
"loss": 0.3896,
|
| 569 |
"step": 80
|
| 570 |
},
|
| 571 |
{
|
| 572 |
+
"epoch": 2.289719626168224,
|
| 573 |
+
"grad_norm": 0.6538497800730606,
|
| 574 |
+
"learning_rate": 1.5240378616267887e-06,
|
| 575 |
+
"loss": 0.3952,
|
| 576 |
"step": 81
|
| 577 |
},
|
| 578 |
{
|
| 579 |
+
"epoch": 2.317757009345794,
|
| 580 |
+
"grad_norm": 0.6004986923958435,
|
| 581 |
+
"learning_rate": 1.4058815806103542e-06,
|
| 582 |
+
"loss": 0.4167,
|
| 583 |
"step": 82
|
| 584 |
},
|
| 585 |
{
|
| 586 |
+
"epoch": 2.3457943925233646,
|
| 587 |
+
"grad_norm": 0.5871010729484297,
|
| 588 |
+
"learning_rate": 1.2917394717602123e-06,
|
| 589 |
+
"loss": 0.3395,
|
| 590 |
"step": 83
|
| 591 |
},
|
| 592 |
{
|
| 593 |
+
"epoch": 2.3738317757009346,
|
| 594 |
+
"grad_norm": 0.6272792477288378,
|
| 595 |
+
"learning_rate": 1.1817390172633402e-06,
|
| 596 |
+
"loss": 0.3955,
|
| 597 |
"step": 84
|
| 598 |
},
|
| 599 |
{
|
| 600 |
+
"epoch": 2.4018691588785046,
|
| 601 |
+
"grad_norm": 0.602988779712659,
|
| 602 |
+
"learning_rate": 1.0760030736066952e-06,
|
| 603 |
+
"loss": 0.3712,
|
| 604 |
"step": 85
|
| 605 |
},
|
| 606 |
{
|
| 607 |
+
"epoch": 2.4299065420560746,
|
| 608 |
+
"grad_norm": 0.6514347127868433,
|
| 609 |
+
"learning_rate": 9.746497343621857e-07,
|
| 610 |
+
"loss": 0.439,
|
| 611 |
"step": 86
|
| 612 |
},
|
| 613 |
{
|
| 614 |
+
"epoch": 2.457943925233645,
|
| 615 |
+
"grad_norm": 0.6171164966676075,
|
| 616 |
+
"learning_rate": 8.777921982911996e-07,
|
| 617 |
+
"loss": 0.3293,
|
| 618 |
"step": 87
|
| 619 |
},
|
| 620 |
{
|
| 621 |
+
"epoch": 2.485981308411215,
|
| 622 |
+
"grad_norm": 0.5985061076335229,
|
| 623 |
+
"learning_rate": 7.85538642916015e-07,
|
| 624 |
+
"loss": 0.3554,
|
| 625 |
"step": 88
|
| 626 |
},
|
| 627 |
{
|
| 628 |
+
"epoch": 2.514018691588785,
|
| 629 |
+
"grad_norm": 0.6023670543052161,
|
| 630 |
+
"learning_rate": 6.979921036993042e-07,
|
| 631 |
+
"loss": 0.3759,
|
| 632 |
"step": 89
|
| 633 |
},
|
| 634 |
{
|
| 635 |
+
"epoch": 2.542056074766355,
|
| 636 |
+
"grad_norm": 0.6350504526148657,
|
| 637 |
+
"learning_rate": 6.152503589666426e-07,
|
| 638 |
+
"loss": 0.4048,
|
| 639 |
"step": 90
|
| 640 |
},
|
| 641 |
{
|
| 642 |
+
"epoch": 2.5700934579439254,
|
| 643 |
+
"grad_norm": 0.6232837840764863,
|
| 644 |
+
"learning_rate": 5.374058207005945e-07,
|
| 645 |
+
"loss": 0.4454,
|
| 646 |
"step": 91
|
| 647 |
},
|
| 648 |
{
|
| 649 |
+
"epoch": 2.5981308411214954,
|
| 650 |
+
"grad_norm": 0.6261955186268706,
|
| 651 |
+
"learning_rate": 4.6454543132829653e-07,
|
| 652 |
+
"loss": 0.3954,
|
| 653 |
"step": 92
|
| 654 |
},
|
| 655 |
{
|
| 656 |
+
"epoch": 2.6261682242990654,
|
| 657 |
+
"grad_norm": 0.6079216763424581,
|
| 658 |
+
"learning_rate": 3.9675056661785563e-07,
|
| 659 |
+
"loss": 0.4013,
|
| 660 |
"step": 93
|
| 661 |
},
|
| 662 |
{
|
| 663 |
+
"epoch": 2.6542056074766354,
|
| 664 |
+
"grad_norm": 0.6437379172509019,
|
| 665 |
+
"learning_rate": 3.340969447919873e-07,
|
| 666 |
+
"loss": 0.4624,
|
| 667 |
"step": 94
|
| 668 |
},
|
| 669 |
{
|
| 670 |
+
"epoch": 2.6822429906542054,
|
| 671 |
+
"grad_norm": 0.6182485342353894,
|
| 672 |
+
"learning_rate": 2.7665454196040665e-07,
|
| 673 |
+
"loss": 0.3614,
|
| 674 |
"step": 95
|
| 675 |
},
|
| 676 |
{
|
| 677 |
+
"epoch": 2.710280373831776,
|
| 678 |
+
"grad_norm": 0.6138213235934379,
|
| 679 |
+
"learning_rate": 2.2448751396543788e-07,
|
| 680 |
+
"loss": 0.3844,
|
| 681 |
"step": 96
|
| 682 |
},
|
| 683 |
{
|
| 684 |
+
"epoch": 2.7383177570093458,
|
| 685 |
+
"grad_norm": 0.5790126523883554,
|
| 686 |
+
"learning_rate": 1.776541247281177e-07,
|
| 687 |
+
"loss": 0.3865,
|
| 688 |
"step": 97
|
| 689 |
},
|
| 690 |
{
|
| 691 |
+
"epoch": 2.7663551401869158,
|
| 692 |
+
"grad_norm": 0.602742444058507,
|
| 693 |
+
"learning_rate": 1.3620668117481471e-07,
|
| 694 |
+
"loss": 0.4459,
|
| 695 |
"step": 98
|
| 696 |
},
|
| 697 |
{
|
| 698 |
+
"epoch": 2.794392523364486,
|
| 699 |
+
"grad_norm": 0.6125089915071265,
|
| 700 |
+
"learning_rate": 1.0019147481706626e-07,
|
| 701 |
+
"loss": 0.4199,
|
| 702 |
"step": 99
|
| 703 |
},
|
| 704 |
{
|
| 705 |
+
"epoch": 2.822429906542056,
|
| 706 |
+
"grad_norm": 0.6081244590976984,
|
| 707 |
+
"learning_rate": 6.964873004985717e-08,
|
| 708 |
+
"loss": 0.4009,
|
| 709 |
"step": 100
|
| 710 |
},
|
| 711 |
{
|
| 712 |
+
"epoch": 2.850467289719626,
|
| 713 |
+
"grad_norm": 0.7145788800241012,
|
| 714 |
+
"learning_rate": 4.461255922609986e-08,
|
| 715 |
+
"loss": 0.3987,
|
| 716 |
"step": 101
|
| 717 |
},
|
| 718 |
{
|
| 719 |
+
"epoch": 2.878504672897196,
|
| 720 |
+
"grad_norm": 0.6581593717216647,
|
| 721 |
+
"learning_rate": 2.511092455747932e-08,
|
| 722 |
+
"loss": 0.3866,
|
| 723 |
"step": 102
|
| 724 |
},
|
| 725 |
{
|
| 726 |
+
"epoch": 2.906542056074766,
|
| 727 |
+
"grad_norm": 0.6290938268876005,
|
| 728 |
+
"learning_rate": 1.1165606884234182e-08,
|
| 729 |
+
"loss": 0.4225,
|
| 730 |
"step": 103
|
| 731 |
},
|
| 732 |
{
|
| 733 |
+
"epoch": 2.9345794392523366,
|
| 734 |
+
"grad_norm": 0.6534972156634564,
|
| 735 |
+
"learning_rate": 2.792181348726941e-09,
|
| 736 |
+
"loss": 0.4279,
|
| 737 |
"step": 104
|
| 738 |
},
|
| 739 |
{
|
| 740 |
+
"epoch": 2.9626168224299065,
|
| 741 |
+
"grad_norm": 0.58156649259114,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 742 |
"learning_rate": 0.0,
|
| 743 |
+
"loss": 0.3581,
|
| 744 |
+
"step": 105
|
| 745 |
},
|
| 746 |
{
|
| 747 |
+
"epoch": 2.9626168224299065,
|
| 748 |
+
"step": 105,
|
| 749 |
+
"total_flos": 18524510396416.0,
|
| 750 |
+
"train_loss": 0.47221575805119104,
|
| 751 |
+
"train_runtime": 641.5687,
|
| 752 |
+
"train_samples_per_second": 15.913,
|
| 753 |
+
"train_steps_per_second": 0.164
|
| 754 |
}
|
| 755 |
],
|
| 756 |
"logging_steps": 1,
|
| 757 |
+
"max_steps": 105,
|
| 758 |
"num_input_tokens_seen": 0,
|
| 759 |
"num_train_epochs": 3,
|
| 760 |
"save_steps": 500,
|
|
|
|
| 770 |
"attributes": {}
|
| 771 |
}
|
| 772 |
},
|
| 773 |
+
"total_flos": 18524510396416.0,
|
| 774 |
+
"train_batch_size": 1,
|
| 775 |
"trial_name": null,
|
| 776 |
"trial_params": null
|
| 777 |
}
|
training_loss.png
CHANGED
|
|