Training in progress, step 32
Browse files- adapter_config.json +4 -4
- adapter_model.safetensors +2 -2
- all_results.json +4 -4
- best_metrics.json +1 -1
- train_results.json +4 -4
- trainer_state.json +262 -262
- training_args.bin +1 -1
adapter_config.json
CHANGED
|
@@ -15,16 +15,16 @@
|
|
| 15 |
"megatron_core": "megatron.core",
|
| 16 |
"modules_to_save": null,
|
| 17 |
"peft_type": "LORA",
|
| 18 |
-
"r":
|
| 19 |
"rank_pattern": {},
|
| 20 |
"revision": null,
|
| 21 |
"target_modules": [
|
| 22 |
-
"
|
| 23 |
"o_proj",
|
| 24 |
"v_proj",
|
| 25 |
-
"
|
| 26 |
-
"up_proj",
|
| 27 |
"gate_proj",
|
|
|
|
| 28 |
"k_proj"
|
| 29 |
],
|
| 30 |
"task_type": "CAUSAL_LM",
|
|
|
|
| 15 |
"megatron_core": "megatron.core",
|
| 16 |
"modules_to_save": null,
|
| 17 |
"peft_type": "LORA",
|
| 18 |
+
"r": 64,
|
| 19 |
"rank_pattern": {},
|
| 20 |
"revision": null,
|
| 21 |
"target_modules": [
|
| 22 |
+
"up_proj",
|
| 23 |
"o_proj",
|
| 24 |
"v_proj",
|
| 25 |
+
"down_proj",
|
|
|
|
| 26 |
"gate_proj",
|
| 27 |
+
"q_proj",
|
| 28 |
"k_proj"
|
| 29 |
],
|
| 30 |
"task_type": "CAUSAL_LM",
|
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 sha256:043c1f9754bae8c5a065227e6b98b91c218b0c06b197b9e0c892fc32c67fb5b9
|
| 3 |
+
size 156927136
|
all_results.json
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
{
|
| 2 |
"epoch": 3.0,
|
| 3 |
-
"train_loss": 0.
|
| 4 |
-
"train_runtime":
|
| 5 |
-
"train_samples_per_second": 7.
|
| 6 |
-
"train_steps_per_second": 0.
|
| 7 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"epoch": 3.0,
|
| 3 |
+
"train_loss": 0.48598923115059733,
|
| 4 |
+
"train_runtime": 1129.8879,
|
| 5 |
+
"train_samples_per_second": 7.25,
|
| 6 |
+
"train_steps_per_second": 0.453
|
| 7 |
}
|
best_metrics.json
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
{"eval_loss": 0.
|
|
|
|
| 1 |
+
{"eval_loss": 0.383393794298172}
|
train_results.json
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
{
|
| 2 |
"epoch": 3.0,
|
| 3 |
-
"train_loss": 0.
|
| 4 |
-
"train_runtime":
|
| 5 |
-
"train_samples_per_second": 7.
|
| 6 |
-
"train_steps_per_second": 0.
|
| 7 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"epoch": 3.0,
|
| 3 |
+
"train_loss": 0.48598923115059733,
|
| 4 |
+
"train_runtime": 1129.8879,
|
| 5 |
+
"train_samples_per_second": 7.25,
|
| 6 |
+
"train_steps_per_second": 0.453
|
| 7 |
}
|
trainer_state.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"best_metric": 0.
|
| 3 |
"best_model_checkpoint": "outputs/checkpoint-512",
|
| 4 |
"epoch": 3.002932551319648,
|
| 5 |
"eval_steps": 32,
|
|
@@ -10,588 +10,588 @@
|
|
| 10 |
"log_history": [
|
| 11 |
{
|
| 12 |
"epoch": 0.05,
|
| 13 |
-
"grad_norm": 0.
|
| 14 |
-
"learning_rate":
|
| 15 |
-
"loss": 1.
|
| 16 |
"step": 8
|
| 17 |
},
|
| 18 |
{
|
| 19 |
"epoch": 0.09,
|
| 20 |
-
"grad_norm": 0.
|
| 21 |
-
"learning_rate": 0.
|
| 22 |
-
"loss": 1.
|
| 23 |
"step": 16
|
| 24 |
},
|
| 25 |
{
|
| 26 |
"epoch": 0.14,
|
| 27 |
-
"grad_norm": 0.
|
| 28 |
-
"learning_rate": 0.
|
| 29 |
-
"loss":
|
| 30 |
"step": 24
|
| 31 |
},
|
| 32 |
{
|
| 33 |
"epoch": 0.19,
|
| 34 |
-
"grad_norm": 0.
|
| 35 |
-
"learning_rate": 0.
|
| 36 |
-
"loss": 0.
|
| 37 |
"step": 32
|
| 38 |
},
|
| 39 |
{
|
| 40 |
"epoch": 0.19,
|
| 41 |
-
"eval_loss": 0.
|
| 42 |
-
"eval_runtime": 2.
|
| 43 |
-
"eval_samples_per_second":
|
| 44 |
-
"eval_steps_per_second": 2.
|
| 45 |
"step": 32
|
| 46 |
},
|
| 47 |
{
|
| 48 |
"epoch": 0.23,
|
| 49 |
-
"grad_norm": 0.
|
| 50 |
-
"learning_rate": 0.
|
| 51 |
-
"loss": 0.
|
| 52 |
"step": 40
|
| 53 |
},
|
| 54 |
{
|
| 55 |
"epoch": 0.28,
|
| 56 |
-
"grad_norm": 0.
|
| 57 |
-
"learning_rate": 0.
|
| 58 |
-
"loss": 0.
|
| 59 |
"step": 48
|
| 60 |
},
|
| 61 |
{
|
| 62 |
"epoch": 0.33,
|
| 63 |
-
"grad_norm": 0.
|
| 64 |
-
"learning_rate": 0.
|
| 65 |
-
"loss": 0.
|
| 66 |
"step": 56
|
| 67 |
},
|
| 68 |
{
|
| 69 |
"epoch": 0.38,
|
| 70 |
-
"grad_norm": 0.
|
| 71 |
-
"learning_rate": 0.
|
| 72 |
-
"loss": 0.
|
| 73 |
"step": 64
|
| 74 |
},
|
| 75 |
{
|
| 76 |
"epoch": 0.38,
|
| 77 |
-
"eval_loss": 0.
|
| 78 |
-
"eval_runtime": 2.
|
| 79 |
-
"eval_samples_per_second":
|
| 80 |
-
"eval_steps_per_second": 2.
|
| 81 |
"step": 64
|
| 82 |
},
|
| 83 |
{
|
| 84 |
"epoch": 0.42,
|
| 85 |
-
"grad_norm": 0.
|
| 86 |
-
"learning_rate": 0.
|
| 87 |
-
"loss": 0.
|
| 88 |
"step": 72
|
| 89 |
},
|
| 90 |
{
|
| 91 |
"epoch": 0.47,
|
| 92 |
-
"grad_norm": 0.
|
| 93 |
-
"learning_rate": 0.
|
| 94 |
-
"loss": 0.
|
| 95 |
"step": 80
|
| 96 |
},
|
| 97 |
{
|
| 98 |
"epoch": 0.52,
|
| 99 |
-
"grad_norm": 0.
|
| 100 |
-
"learning_rate": 0.
|
| 101 |
-
"loss": 0.
|
| 102 |
"step": 88
|
| 103 |
},
|
| 104 |
{
|
| 105 |
"epoch": 0.56,
|
| 106 |
-
"grad_norm": 0.
|
| 107 |
-
"learning_rate": 0.
|
| 108 |
-
"loss": 0.
|
| 109 |
"step": 96
|
| 110 |
},
|
| 111 |
{
|
| 112 |
"epoch": 0.56,
|
| 113 |
-
"eval_loss": 0.
|
| 114 |
-
"eval_runtime": 2.
|
| 115 |
-
"eval_samples_per_second":
|
| 116 |
-
"eval_steps_per_second": 2.
|
| 117 |
"step": 96
|
| 118 |
},
|
| 119 |
{
|
| 120 |
"epoch": 0.61,
|
| 121 |
-
"grad_norm": 0.
|
| 122 |
-
"learning_rate": 0.
|
| 123 |
-
"loss": 0.
|
| 124 |
"step": 104
|
| 125 |
},
|
| 126 |
{
|
| 127 |
"epoch": 0.66,
|
| 128 |
-
"grad_norm": 0.
|
| 129 |
-
"learning_rate": 0.
|
| 130 |
-
"loss": 0.
|
| 131 |
"step": 112
|
| 132 |
},
|
| 133 |
{
|
| 134 |
"epoch": 0.7,
|
| 135 |
-
"grad_norm": 0.
|
| 136 |
-
"learning_rate": 0.
|
| 137 |
-
"loss": 0.
|
| 138 |
"step": 120
|
| 139 |
},
|
| 140 |
{
|
| 141 |
"epoch": 0.75,
|
| 142 |
-
"grad_norm": 0.
|
| 143 |
-
"learning_rate": 0.
|
| 144 |
-
"loss": 0.
|
| 145 |
"step": 128
|
| 146 |
},
|
| 147 |
{
|
| 148 |
"epoch": 0.75,
|
| 149 |
-
"eval_loss": 0.
|
| 150 |
-
"eval_runtime": 2.
|
| 151 |
-
"eval_samples_per_second":
|
| 152 |
-
"eval_steps_per_second": 2.
|
| 153 |
"step": 128
|
| 154 |
},
|
| 155 |
{
|
| 156 |
"epoch": 0.8,
|
| 157 |
-
"grad_norm": 0.
|
| 158 |
-
"learning_rate": 0.
|
| 159 |
-
"loss": 0.
|
| 160 |
"step": 136
|
| 161 |
},
|
| 162 |
{
|
| 163 |
"epoch": 0.84,
|
| 164 |
-
"grad_norm": 0.
|
| 165 |
-
"learning_rate": 0.
|
| 166 |
-
"loss": 0.
|
| 167 |
"step": 144
|
| 168 |
},
|
| 169 |
{
|
| 170 |
"epoch": 0.89,
|
| 171 |
-
"grad_norm": 0.
|
| 172 |
-
"learning_rate": 0.
|
| 173 |
-
"loss": 0.
|
| 174 |
"step": 152
|
| 175 |
},
|
| 176 |
{
|
| 177 |
"epoch": 0.94,
|
| 178 |
-
"grad_norm": 0.
|
| 179 |
-
"learning_rate": 0.
|
| 180 |
-
"loss": 0.
|
| 181 |
"step": 160
|
| 182 |
},
|
| 183 |
{
|
| 184 |
"epoch": 0.94,
|
| 185 |
-
"eval_loss": 0.
|
| 186 |
-
"eval_runtime": 2.
|
| 187 |
-
"eval_samples_per_second":
|
| 188 |
-
"eval_steps_per_second": 2.
|
| 189 |
"step": 160
|
| 190 |
},
|
| 191 |
{
|
| 192 |
"epoch": 0.99,
|
| 193 |
-
"grad_norm": 0.
|
| 194 |
-
"learning_rate": 0.
|
| 195 |
-
"loss": 0.
|
| 196 |
"step": 168
|
| 197 |
},
|
| 198 |
{
|
| 199 |
"epoch": 1.03,
|
| 200 |
-
"grad_norm": 0.
|
| 201 |
-
"learning_rate": 0.
|
| 202 |
-
"loss": 0.
|
| 203 |
"step": 176
|
| 204 |
},
|
| 205 |
{
|
| 206 |
"epoch": 1.08,
|
| 207 |
-
"grad_norm": 0.
|
| 208 |
-
"learning_rate": 0.
|
| 209 |
-
"loss": 0.
|
| 210 |
"step": 184
|
| 211 |
},
|
| 212 |
{
|
| 213 |
"epoch": 1.13,
|
| 214 |
-
"grad_norm": 0.
|
| 215 |
-
"learning_rate": 0.
|
| 216 |
-
"loss": 0.
|
| 217 |
"step": 192
|
| 218 |
},
|
| 219 |
{
|
| 220 |
"epoch": 1.13,
|
| 221 |
-
"eval_loss": 0.
|
| 222 |
-
"eval_runtime": 2.
|
| 223 |
-
"eval_samples_per_second":
|
| 224 |
-
"eval_steps_per_second": 2.
|
| 225 |
"step": 192
|
| 226 |
},
|
| 227 |
{
|
| 228 |
"epoch": 1.17,
|
| 229 |
-
"grad_norm": 0.
|
| 230 |
-
"learning_rate": 0.
|
| 231 |
-
"loss": 0.
|
| 232 |
"step": 200
|
| 233 |
},
|
| 234 |
{
|
| 235 |
"epoch": 1.22,
|
| 236 |
-
"grad_norm": 0.
|
| 237 |
-
"learning_rate": 0.
|
| 238 |
-
"loss": 0.
|
| 239 |
"step": 208
|
| 240 |
},
|
| 241 |
{
|
| 242 |
"epoch": 1.27,
|
| 243 |
-
"grad_norm": 0.
|
| 244 |
-
"learning_rate": 0.
|
| 245 |
-
"loss": 0.
|
| 246 |
"step": 216
|
| 247 |
},
|
| 248 |
{
|
| 249 |
"epoch": 1.31,
|
| 250 |
-
"grad_norm": 0.
|
| 251 |
-
"learning_rate": 0.
|
| 252 |
-
"loss": 0.
|
| 253 |
"step": 224
|
| 254 |
},
|
| 255 |
{
|
| 256 |
"epoch": 1.31,
|
| 257 |
-
"eval_loss": 0.
|
| 258 |
-
"eval_runtime": 2.
|
| 259 |
-
"eval_samples_per_second":
|
| 260 |
-
"eval_steps_per_second": 2.
|
| 261 |
"step": 224
|
| 262 |
},
|
| 263 |
{
|
| 264 |
"epoch": 1.36,
|
| 265 |
-
"grad_norm": 0.
|
| 266 |
-
"learning_rate": 0.
|
| 267 |
-
"loss": 0.
|
| 268 |
"step": 232
|
| 269 |
},
|
| 270 |
{
|
| 271 |
"epoch": 1.41,
|
| 272 |
-
"grad_norm": 0.
|
| 273 |
-
"learning_rate": 0.
|
| 274 |
-
"loss": 0.
|
| 275 |
"step": 240
|
| 276 |
},
|
| 277 |
{
|
| 278 |
"epoch": 1.45,
|
| 279 |
-
"grad_norm": 0.
|
| 280 |
-
"learning_rate": 0.
|
| 281 |
-
"loss": 0.
|
| 282 |
"step": 248
|
| 283 |
},
|
| 284 |
{
|
| 285 |
"epoch": 1.5,
|
| 286 |
-
"grad_norm": 0.
|
| 287 |
-
"learning_rate": 0.
|
| 288 |
-
"loss": 0.
|
| 289 |
"step": 256
|
| 290 |
},
|
| 291 |
{
|
| 292 |
"epoch": 1.5,
|
| 293 |
-
"eval_loss": 0.
|
| 294 |
-
"eval_runtime": 2.
|
| 295 |
-
"eval_samples_per_second":
|
| 296 |
-
"eval_steps_per_second": 2.
|
| 297 |
"step": 256
|
| 298 |
},
|
| 299 |
{
|
| 300 |
"epoch": 1.55,
|
| 301 |
-
"grad_norm": 0.
|
| 302 |
-
"learning_rate": 0.
|
| 303 |
-
"loss": 0.
|
| 304 |
"step": 264
|
| 305 |
},
|
| 306 |
{
|
| 307 |
"epoch": 1.6,
|
| 308 |
-
"grad_norm": 0.
|
| 309 |
-
"learning_rate": 0.
|
| 310 |
-
"loss": 0.
|
| 311 |
"step": 272
|
| 312 |
},
|
| 313 |
{
|
| 314 |
"epoch": 1.64,
|
| 315 |
-
"grad_norm": 0.
|
| 316 |
-
"learning_rate":
|
| 317 |
-
"loss": 0.
|
| 318 |
"step": 280
|
| 319 |
},
|
| 320 |
{
|
| 321 |
"epoch": 1.69,
|
| 322 |
-
"grad_norm": 0.
|
| 323 |
-
"learning_rate":
|
| 324 |
-
"loss": 0.
|
| 325 |
"step": 288
|
| 326 |
},
|
| 327 |
{
|
| 328 |
"epoch": 1.69,
|
| 329 |
-
"eval_loss": 0.
|
| 330 |
-
"eval_runtime": 2.
|
| 331 |
-
"eval_samples_per_second":
|
| 332 |
-
"eval_steps_per_second": 2.
|
| 333 |
"step": 288
|
| 334 |
},
|
| 335 |
{
|
| 336 |
"epoch": 1.74,
|
| 337 |
-
"grad_norm": 0.
|
| 338 |
-
"learning_rate":
|
| 339 |
-
"loss": 0.
|
| 340 |
"step": 296
|
| 341 |
},
|
| 342 |
{
|
| 343 |
"epoch": 1.78,
|
| 344 |
-
"grad_norm": 0.
|
| 345 |
-
"learning_rate":
|
| 346 |
-
"loss": 0.
|
| 347 |
"step": 304
|
| 348 |
},
|
| 349 |
{
|
| 350 |
"epoch": 1.83,
|
| 351 |
-
"grad_norm": 0.
|
| 352 |
-
"learning_rate":
|
| 353 |
-
"loss": 0.
|
| 354 |
"step": 312
|
| 355 |
},
|
| 356 |
{
|
| 357 |
"epoch": 1.88,
|
| 358 |
-
"grad_norm": 0.
|
| 359 |
-
"learning_rate":
|
| 360 |
-
"loss": 0.
|
| 361 |
"step": 320
|
| 362 |
},
|
| 363 |
{
|
| 364 |
"epoch": 1.88,
|
| 365 |
-
"eval_loss": 0.
|
| 366 |
-
"eval_runtime": 2.
|
| 367 |
-
"eval_samples_per_second":
|
| 368 |
-
"eval_steps_per_second": 2.
|
| 369 |
"step": 320
|
| 370 |
},
|
| 371 |
{
|
| 372 |
"epoch": 1.92,
|
| 373 |
-
"grad_norm": 0.
|
| 374 |
-
"learning_rate":
|
| 375 |
-
"loss": 0.
|
| 376 |
"step": 328
|
| 377 |
},
|
| 378 |
{
|
| 379 |
"epoch": 1.97,
|
| 380 |
-
"grad_norm": 0.
|
| 381 |
-
"learning_rate":
|
| 382 |
-
"loss": 0.
|
| 383 |
"step": 336
|
| 384 |
},
|
| 385 |
{
|
| 386 |
"epoch": 2.02,
|
| 387 |
-
"grad_norm": 0.
|
| 388 |
-
"learning_rate":
|
| 389 |
-
"loss": 0.
|
| 390 |
"step": 344
|
| 391 |
},
|
| 392 |
{
|
| 393 |
"epoch": 2.06,
|
| 394 |
-
"grad_norm": 0.
|
| 395 |
-
"learning_rate":
|
| 396 |
-
"loss": 0.
|
| 397 |
"step": 352
|
| 398 |
},
|
| 399 |
{
|
| 400 |
"epoch": 2.06,
|
| 401 |
-
"eval_loss": 0.
|
| 402 |
-
"eval_runtime": 2.
|
| 403 |
-
"eval_samples_per_second":
|
| 404 |
-
"eval_steps_per_second": 2.
|
| 405 |
"step": 352
|
| 406 |
},
|
| 407 |
{
|
| 408 |
"epoch": 2.11,
|
| 409 |
-
"grad_norm": 0.
|
| 410 |
-
"learning_rate":
|
| 411 |
-
"loss": 0.
|
| 412 |
"step": 360
|
| 413 |
},
|
| 414 |
{
|
| 415 |
"epoch": 2.16,
|
| 416 |
-
"grad_norm": 0.
|
| 417 |
-
"learning_rate":
|
| 418 |
-
"loss": 0.
|
| 419 |
"step": 368
|
| 420 |
},
|
| 421 |
{
|
| 422 |
"epoch": 2.21,
|
| 423 |
-
"grad_norm": 0.
|
| 424 |
-
"learning_rate":
|
| 425 |
-
"loss": 0.
|
| 426 |
"step": 376
|
| 427 |
},
|
| 428 |
{
|
| 429 |
"epoch": 2.25,
|
| 430 |
-
"grad_norm": 0.
|
| 431 |
-
"learning_rate":
|
| 432 |
-
"loss": 0.
|
| 433 |
"step": 384
|
| 434 |
},
|
| 435 |
{
|
| 436 |
"epoch": 2.25,
|
| 437 |
-
"eval_loss": 0.
|
| 438 |
-
"eval_runtime": 2.
|
| 439 |
-
"eval_samples_per_second":
|
| 440 |
-
"eval_steps_per_second": 2.
|
| 441 |
"step": 384
|
| 442 |
},
|
| 443 |
{
|
| 444 |
"epoch": 2.3,
|
| 445 |
-
"grad_norm": 0.
|
| 446 |
-
"learning_rate":
|
| 447 |
-
"loss": 0.
|
| 448 |
"step": 392
|
| 449 |
},
|
| 450 |
{
|
| 451 |
"epoch": 2.35,
|
| 452 |
-
"grad_norm": 0.
|
| 453 |
-
"learning_rate":
|
| 454 |
-
"loss": 0.
|
| 455 |
"step": 400
|
| 456 |
},
|
| 457 |
{
|
| 458 |
"epoch": 2.39,
|
| 459 |
-
"grad_norm": 0.
|
| 460 |
-
"learning_rate":
|
| 461 |
-
"loss": 0.
|
| 462 |
"step": 408
|
| 463 |
},
|
| 464 |
{
|
| 465 |
"epoch": 2.44,
|
| 466 |
-
"grad_norm": 0.
|
| 467 |
-
"learning_rate":
|
| 468 |
-
"loss": 0.
|
| 469 |
"step": 416
|
| 470 |
},
|
| 471 |
{
|
| 472 |
"epoch": 2.44,
|
| 473 |
-
"eval_loss": 0.
|
| 474 |
-
"eval_runtime": 2.
|
| 475 |
-
"eval_samples_per_second":
|
| 476 |
-
"eval_steps_per_second": 2.
|
| 477 |
"step": 416
|
| 478 |
},
|
| 479 |
{
|
| 480 |
"epoch": 2.49,
|
| 481 |
-
"grad_norm": 0.
|
| 482 |
-
"learning_rate":
|
| 483 |
-
"loss": 0.
|
| 484 |
"step": 424
|
| 485 |
},
|
| 486 |
{
|
| 487 |
"epoch": 2.53,
|
| 488 |
-
"grad_norm": 0.
|
| 489 |
-
"learning_rate":
|
| 490 |
-
"loss": 0.
|
| 491 |
"step": 432
|
| 492 |
},
|
| 493 |
{
|
| 494 |
"epoch": 2.58,
|
| 495 |
-
"grad_norm": 0.
|
| 496 |
-
"learning_rate":
|
| 497 |
-
"loss": 0.
|
| 498 |
"step": 440
|
| 499 |
},
|
| 500 |
{
|
| 501 |
"epoch": 2.63,
|
| 502 |
-
"grad_norm": 0.
|
| 503 |
-
"learning_rate":
|
| 504 |
-
"loss": 0.
|
| 505 |
"step": 448
|
| 506 |
},
|
| 507 |
{
|
| 508 |
"epoch": 2.63,
|
| 509 |
-
"eval_loss": 0.
|
| 510 |
-
"eval_runtime": 2.
|
| 511 |
-
"eval_samples_per_second":
|
| 512 |
-
"eval_steps_per_second": 2.
|
| 513 |
"step": 448
|
| 514 |
},
|
| 515 |
{
|
| 516 |
"epoch": 2.67,
|
| 517 |
-
"grad_norm": 0.
|
| 518 |
-
"learning_rate":
|
| 519 |
-
"loss": 0.
|
| 520 |
"step": 456
|
| 521 |
},
|
| 522 |
{
|
| 523 |
"epoch": 2.72,
|
| 524 |
-
"grad_norm": 0.
|
| 525 |
-
"learning_rate":
|
| 526 |
-
"loss": 0.
|
| 527 |
"step": 464
|
| 528 |
},
|
| 529 |
{
|
| 530 |
"epoch": 2.77,
|
| 531 |
-
"grad_norm": 0.
|
| 532 |
-
"learning_rate":
|
| 533 |
-
"loss": 0.
|
| 534 |
"step": 472
|
| 535 |
},
|
| 536 |
{
|
| 537 |
"epoch": 2.82,
|
| 538 |
-
"grad_norm": 0.
|
| 539 |
-
"learning_rate":
|
| 540 |
-
"loss": 0.
|
| 541 |
"step": 480
|
| 542 |
},
|
| 543 |
{
|
| 544 |
"epoch": 2.82,
|
| 545 |
-
"eval_loss": 0.
|
| 546 |
-
"eval_runtime": 2.
|
| 547 |
-
"eval_samples_per_second":
|
| 548 |
-
"eval_steps_per_second": 2.
|
| 549 |
"step": 480
|
| 550 |
},
|
| 551 |
{
|
| 552 |
"epoch": 2.86,
|
| 553 |
-
"grad_norm": 0.
|
| 554 |
-
"learning_rate":
|
| 555 |
-
"loss": 0.
|
| 556 |
"step": 488
|
| 557 |
},
|
| 558 |
{
|
| 559 |
"epoch": 2.91,
|
| 560 |
-
"grad_norm": 0.
|
| 561 |
-
"learning_rate":
|
| 562 |
-
"loss": 0.
|
| 563 |
"step": 496
|
| 564 |
},
|
| 565 |
{
|
| 566 |
"epoch": 2.96,
|
| 567 |
-
"grad_norm": 0.
|
| 568 |
-
"learning_rate":
|
| 569 |
-
"loss": 0.
|
| 570 |
"step": 504
|
| 571 |
},
|
| 572 |
{
|
| 573 |
"epoch": 3.0,
|
| 574 |
-
"grad_norm": 0.
|
| 575 |
"learning_rate": 0.0,
|
| 576 |
-
"loss": 0.
|
| 577 |
"step": 512
|
| 578 |
},
|
| 579 |
{
|
| 580 |
"epoch": 3.0,
|
| 581 |
-
"eval_loss": 0.
|
| 582 |
-
"eval_runtime": 2.
|
| 583 |
-
"eval_samples_per_second": 21.
|
| 584 |
-
"eval_steps_per_second": 2.
|
| 585 |
"step": 512
|
| 586 |
},
|
| 587 |
{
|
| 588 |
"epoch": 3.0,
|
| 589 |
"step": 512,
|
| 590 |
-
"total_flos":
|
| 591 |
-
"train_loss": 0.
|
| 592 |
-
"train_runtime":
|
| 593 |
-
"train_samples_per_second": 7.
|
| 594 |
-
"train_steps_per_second": 0.
|
| 595 |
}
|
| 596 |
],
|
| 597 |
"logging_steps": 8,
|
|
@@ -599,7 +599,7 @@
|
|
| 599 |
"num_input_tokens_seen": 0,
|
| 600 |
"num_train_epochs": 4,
|
| 601 |
"save_steps": 32,
|
| 602 |
-
"total_flos":
|
| 603 |
"train_batch_size": 4,
|
| 604 |
"trial_name": null,
|
| 605 |
"trial_params": null
|
|
|
|
| 1 |
{
|
| 2 |
+
"best_metric": 0.383393794298172,
|
| 3 |
"best_model_checkpoint": "outputs/checkpoint-512",
|
| 4 |
"epoch": 3.002932551319648,
|
| 5 |
"eval_steps": 32,
|
|
|
|
| 10 |
"log_history": [
|
| 11 |
{
|
| 12 |
"epoch": 0.05,
|
| 13 |
+
"grad_norm": 0.125,
|
| 14 |
+
"learning_rate": 0.000125,
|
| 15 |
+
"loss": 1.2564,
|
| 16 |
"step": 8
|
| 17 |
},
|
| 18 |
{
|
| 19 |
"epoch": 0.09,
|
| 20 |
+
"grad_norm": 0.1337890625,
|
| 21 |
+
"learning_rate": 0.00025,
|
| 22 |
+
"loss": 1.1071,
|
| 23 |
"step": 16
|
| 24 |
},
|
| 25 |
{
|
| 26 |
"epoch": 0.14,
|
| 27 |
+
"grad_norm": 0.1181640625,
|
| 28 |
+
"learning_rate": 0.000375,
|
| 29 |
+
"loss": 0.8351,
|
| 30 |
"step": 24
|
| 31 |
},
|
| 32 |
{
|
| 33 |
"epoch": 0.19,
|
| 34 |
+
"grad_norm": 0.11181640625,
|
| 35 |
+
"learning_rate": 0.0005,
|
| 36 |
+
"loss": 0.709,
|
| 37 |
"step": 32
|
| 38 |
},
|
| 39 |
{
|
| 40 |
"epoch": 0.19,
|
| 41 |
+
"eval_loss": 0.5833926796913147,
|
| 42 |
+
"eval_runtime": 2.7901,
|
| 43 |
+
"eval_samples_per_second": 17.203,
|
| 44 |
+
"eval_steps_per_second": 2.15,
|
| 45 |
"step": 32
|
| 46 |
},
|
| 47 |
{
|
| 48 |
"epoch": 0.23,
|
| 49 |
+
"grad_norm": 0.1064453125,
|
| 50 |
+
"learning_rate": 0.0004916666666666666,
|
| 51 |
+
"loss": 0.6104,
|
| 52 |
"step": 40
|
| 53 |
},
|
| 54 |
{
|
| 55 |
"epoch": 0.28,
|
| 56 |
+
"grad_norm": 0.10693359375,
|
| 57 |
+
"learning_rate": 0.00048333333333333334,
|
| 58 |
+
"loss": 0.5962,
|
| 59 |
"step": 48
|
| 60 |
},
|
| 61 |
{
|
| 62 |
"epoch": 0.33,
|
| 63 |
+
"grad_norm": 0.099609375,
|
| 64 |
+
"learning_rate": 0.000475,
|
| 65 |
+
"loss": 0.5696,
|
| 66 |
"step": 56
|
| 67 |
},
|
| 68 |
{
|
| 69 |
"epoch": 0.38,
|
| 70 |
+
"grad_norm": 0.13671875,
|
| 71 |
+
"learning_rate": 0.00046666666666666666,
|
| 72 |
+
"loss": 0.5678,
|
| 73 |
"step": 64
|
| 74 |
},
|
| 75 |
{
|
| 76 |
"epoch": 0.38,
|
| 77 |
+
"eval_loss": 0.47890207171440125,
|
| 78 |
+
"eval_runtime": 2.176,
|
| 79 |
+
"eval_samples_per_second": 22.059,
|
| 80 |
+
"eval_steps_per_second": 2.757,
|
| 81 |
"step": 64
|
| 82 |
},
|
| 83 |
{
|
| 84 |
"epoch": 0.42,
|
| 85 |
+
"grad_norm": 0.0849609375,
|
| 86 |
+
"learning_rate": 0.0004583333333333333,
|
| 87 |
+
"loss": 0.5528,
|
| 88 |
"step": 72
|
| 89 |
},
|
| 90 |
{
|
| 91 |
"epoch": 0.47,
|
| 92 |
+
"grad_norm": 0.0810546875,
|
| 93 |
+
"learning_rate": 0.00045000000000000004,
|
| 94 |
+
"loss": 0.535,
|
| 95 |
"step": 80
|
| 96 |
},
|
| 97 |
{
|
| 98 |
"epoch": 0.52,
|
| 99 |
+
"grad_norm": 0.0771484375,
|
| 100 |
+
"learning_rate": 0.00044166666666666665,
|
| 101 |
+
"loss": 0.5049,
|
| 102 |
"step": 88
|
| 103 |
},
|
| 104 |
{
|
| 105 |
"epoch": 0.56,
|
| 106 |
+
"grad_norm": 0.07275390625,
|
| 107 |
+
"learning_rate": 0.00043333333333333337,
|
| 108 |
+
"loss": 0.5165,
|
| 109 |
"step": 96
|
| 110 |
},
|
| 111 |
{
|
| 112 |
"epoch": 0.56,
|
| 113 |
+
"eval_loss": 0.44684532284736633,
|
| 114 |
+
"eval_runtime": 2.1779,
|
| 115 |
+
"eval_samples_per_second": 22.04,
|
| 116 |
+
"eval_steps_per_second": 2.755,
|
| 117 |
"step": 96
|
| 118 |
},
|
| 119 |
{
|
| 120 |
"epoch": 0.61,
|
| 121 |
+
"grad_norm": 0.068359375,
|
| 122 |
+
"learning_rate": 0.000425,
|
| 123 |
+
"loss": 0.5133,
|
| 124 |
"step": 104
|
| 125 |
},
|
| 126 |
{
|
| 127 |
"epoch": 0.66,
|
| 128 |
+
"grad_norm": 0.076171875,
|
| 129 |
+
"learning_rate": 0.0004166666666666667,
|
| 130 |
+
"loss": 0.4931,
|
| 131 |
"step": 112
|
| 132 |
},
|
| 133 |
{
|
| 134 |
"epoch": 0.7,
|
| 135 |
+
"grad_norm": 0.07373046875,
|
| 136 |
+
"learning_rate": 0.00040833333333333336,
|
| 137 |
+
"loss": 0.5168,
|
| 138 |
"step": 120
|
| 139 |
},
|
| 140 |
{
|
| 141 |
"epoch": 0.75,
|
| 142 |
+
"grad_norm": 0.06787109375,
|
| 143 |
+
"learning_rate": 0.0004,
|
| 144 |
+
"loss": 0.4926,
|
| 145 |
"step": 128
|
| 146 |
},
|
| 147 |
{
|
| 148 |
"epoch": 0.75,
|
| 149 |
+
"eval_loss": 0.42885932326316833,
|
| 150 |
+
"eval_runtime": 2.1761,
|
| 151 |
+
"eval_samples_per_second": 22.058,
|
| 152 |
+
"eval_steps_per_second": 2.757,
|
| 153 |
"step": 128
|
| 154 |
},
|
| 155 |
{
|
| 156 |
"epoch": 0.8,
|
| 157 |
+
"grad_norm": 0.07080078125,
|
| 158 |
+
"learning_rate": 0.0003916666666666667,
|
| 159 |
+
"loss": 0.4843,
|
| 160 |
"step": 136
|
| 161 |
},
|
| 162 |
{
|
| 163 |
"epoch": 0.84,
|
| 164 |
+
"grad_norm": 0.0712890625,
|
| 165 |
+
"learning_rate": 0.00038333333333333334,
|
| 166 |
+
"loss": 0.4697,
|
| 167 |
"step": 144
|
| 168 |
},
|
| 169 |
{
|
| 170 |
"epoch": 0.89,
|
| 171 |
+
"grad_norm": 0.08056640625,
|
| 172 |
+
"learning_rate": 0.000375,
|
| 173 |
+
"loss": 0.4681,
|
| 174 |
"step": 152
|
| 175 |
},
|
| 176 |
{
|
| 177 |
"epoch": 0.94,
|
| 178 |
+
"grad_norm": 0.0732421875,
|
| 179 |
+
"learning_rate": 0.00036666666666666667,
|
| 180 |
+
"loss": 0.4667,
|
| 181 |
"step": 160
|
| 182 |
},
|
| 183 |
{
|
| 184 |
"epoch": 0.94,
|
| 185 |
+
"eval_loss": 0.41727420687675476,
|
| 186 |
+
"eval_runtime": 2.1762,
|
| 187 |
+
"eval_samples_per_second": 22.057,
|
| 188 |
+
"eval_steps_per_second": 2.757,
|
| 189 |
"step": 160
|
| 190 |
},
|
| 191 |
{
|
| 192 |
"epoch": 0.99,
|
| 193 |
+
"grad_norm": 0.07177734375,
|
| 194 |
+
"learning_rate": 0.00035833333333333333,
|
| 195 |
+
"loss": 0.4721,
|
| 196 |
"step": 168
|
| 197 |
},
|
| 198 |
{
|
| 199 |
"epoch": 1.03,
|
| 200 |
+
"grad_norm": 0.08349609375,
|
| 201 |
+
"learning_rate": 0.00035,
|
| 202 |
+
"loss": 0.471,
|
| 203 |
"step": 176
|
| 204 |
},
|
| 205 |
{
|
| 206 |
"epoch": 1.08,
|
| 207 |
+
"grad_norm": 0.07421875,
|
| 208 |
+
"learning_rate": 0.00034166666666666666,
|
| 209 |
+
"loss": 0.4544,
|
| 210 |
"step": 184
|
| 211 |
},
|
| 212 |
{
|
| 213 |
"epoch": 1.13,
|
| 214 |
+
"grad_norm": 0.076171875,
|
| 215 |
+
"learning_rate": 0.0003333333333333333,
|
| 216 |
+
"loss": 0.443,
|
| 217 |
"step": 192
|
| 218 |
},
|
| 219 |
{
|
| 220 |
"epoch": 1.13,
|
| 221 |
+
"eval_loss": 0.4128858149051666,
|
| 222 |
+
"eval_runtime": 2.1785,
|
| 223 |
+
"eval_samples_per_second": 22.033,
|
| 224 |
+
"eval_steps_per_second": 2.754,
|
| 225 |
"step": 192
|
| 226 |
},
|
| 227 |
{
|
| 228 |
"epoch": 1.17,
|
| 229 |
+
"grad_norm": 0.08251953125,
|
| 230 |
+
"learning_rate": 0.00032500000000000004,
|
| 231 |
+
"loss": 0.4423,
|
| 232 |
"step": 200
|
| 233 |
},
|
| 234 |
{
|
| 235 |
"epoch": 1.22,
|
| 236 |
+
"grad_norm": 0.0869140625,
|
| 237 |
+
"learning_rate": 0.00031666666666666665,
|
| 238 |
+
"loss": 0.4704,
|
| 239 |
"step": 208
|
| 240 |
},
|
| 241 |
{
|
| 242 |
"epoch": 1.27,
|
| 243 |
+
"grad_norm": 0.07763671875,
|
| 244 |
+
"learning_rate": 0.00030833333333333337,
|
| 245 |
+
"loss": 0.4449,
|
| 246 |
"step": 216
|
| 247 |
},
|
| 248 |
{
|
| 249 |
"epoch": 1.31,
|
| 250 |
+
"grad_norm": 0.08349609375,
|
| 251 |
+
"learning_rate": 0.0003,
|
| 252 |
+
"loss": 0.4313,
|
| 253 |
"step": 224
|
| 254 |
},
|
| 255 |
{
|
| 256 |
"epoch": 1.31,
|
| 257 |
+
"eval_loss": 0.4045116901397705,
|
| 258 |
+
"eval_runtime": 2.1781,
|
| 259 |
+
"eval_samples_per_second": 22.038,
|
| 260 |
+
"eval_steps_per_second": 2.755,
|
| 261 |
"step": 224
|
| 262 |
},
|
| 263 |
{
|
| 264 |
"epoch": 1.36,
|
| 265 |
+
"grad_norm": 0.07666015625,
|
| 266 |
+
"learning_rate": 0.0002916666666666667,
|
| 267 |
+
"loss": 0.4619,
|
| 268 |
"step": 232
|
| 269 |
},
|
| 270 |
{
|
| 271 |
"epoch": 1.41,
|
| 272 |
+
"grad_norm": 0.08154296875,
|
| 273 |
+
"learning_rate": 0.00028333333333333335,
|
| 274 |
+
"loss": 0.4435,
|
| 275 |
"step": 240
|
| 276 |
},
|
| 277 |
{
|
| 278 |
"epoch": 1.45,
|
| 279 |
+
"grad_norm": 0.07763671875,
|
| 280 |
+
"learning_rate": 0.000275,
|
| 281 |
+
"loss": 0.4272,
|
| 282 |
"step": 248
|
| 283 |
},
|
| 284 |
{
|
| 285 |
"epoch": 1.5,
|
| 286 |
+
"grad_norm": 0.083984375,
|
| 287 |
+
"learning_rate": 0.0002666666666666667,
|
| 288 |
+
"loss": 0.4554,
|
| 289 |
"step": 256
|
| 290 |
},
|
| 291 |
{
|
| 292 |
"epoch": 1.5,
|
| 293 |
+
"eval_loss": 0.39897629618644714,
|
| 294 |
+
"eval_runtime": 2.177,
|
| 295 |
+
"eval_samples_per_second": 22.049,
|
| 296 |
+
"eval_steps_per_second": 2.756,
|
| 297 |
"step": 256
|
| 298 |
},
|
| 299 |
{
|
| 300 |
"epoch": 1.55,
|
| 301 |
+
"grad_norm": 0.07373046875,
|
| 302 |
+
"learning_rate": 0.00025833333333333334,
|
| 303 |
+
"loss": 0.438,
|
| 304 |
"step": 264
|
| 305 |
},
|
| 306 |
{
|
| 307 |
"epoch": 1.6,
|
| 308 |
+
"grad_norm": 0.08544921875,
|
| 309 |
+
"learning_rate": 0.00025,
|
| 310 |
+
"loss": 0.438,
|
| 311 |
"step": 272
|
| 312 |
},
|
| 313 |
{
|
| 314 |
"epoch": 1.64,
|
| 315 |
+
"grad_norm": 0.0810546875,
|
| 316 |
+
"learning_rate": 0.00024166666666666667,
|
| 317 |
+
"loss": 0.4332,
|
| 318 |
"step": 280
|
| 319 |
},
|
| 320 |
{
|
| 321 |
"epoch": 1.69,
|
| 322 |
+
"grad_norm": 0.0810546875,
|
| 323 |
+
"learning_rate": 0.00023333333333333333,
|
| 324 |
+
"loss": 0.4453,
|
| 325 |
"step": 288
|
| 326 |
},
|
| 327 |
{
|
| 328 |
"epoch": 1.69,
|
| 329 |
+
"eval_loss": 0.39246222376823425,
|
| 330 |
+
"eval_runtime": 2.1791,
|
| 331 |
+
"eval_samples_per_second": 22.027,
|
| 332 |
+
"eval_steps_per_second": 2.753,
|
| 333 |
"step": 288
|
| 334 |
},
|
| 335 |
{
|
| 336 |
"epoch": 1.74,
|
| 337 |
+
"grad_norm": 0.07080078125,
|
| 338 |
+
"learning_rate": 0.00022500000000000002,
|
| 339 |
+
"loss": 0.468,
|
| 340 |
"step": 296
|
| 341 |
},
|
| 342 |
{
|
| 343 |
"epoch": 1.78,
|
| 344 |
+
"grad_norm": 0.09423828125,
|
| 345 |
+
"learning_rate": 0.00021666666666666668,
|
| 346 |
+
"loss": 0.4311,
|
| 347 |
"step": 304
|
| 348 |
},
|
| 349 |
{
|
| 350 |
"epoch": 1.83,
|
| 351 |
+
"grad_norm": 0.07373046875,
|
| 352 |
+
"learning_rate": 0.00020833333333333335,
|
| 353 |
+
"loss": 0.4537,
|
| 354 |
"step": 312
|
| 355 |
},
|
| 356 |
{
|
| 357 |
"epoch": 1.88,
|
| 358 |
+
"grad_norm": 0.08984375,
|
| 359 |
+
"learning_rate": 0.0002,
|
| 360 |
+
"loss": 0.4369,
|
| 361 |
"step": 320
|
| 362 |
},
|
| 363 |
{
|
| 364 |
"epoch": 1.88,
|
| 365 |
+
"eval_loss": 0.3888448178768158,
|
| 366 |
+
"eval_runtime": 2.1807,
|
| 367 |
+
"eval_samples_per_second": 22.011,
|
| 368 |
+
"eval_steps_per_second": 2.751,
|
| 369 |
"step": 320
|
| 370 |
},
|
| 371 |
{
|
| 372 |
"epoch": 1.92,
|
| 373 |
+
"grad_norm": 0.0693359375,
|
| 374 |
+
"learning_rate": 0.00019166666666666667,
|
| 375 |
+
"loss": 0.443,
|
| 376 |
"step": 328
|
| 377 |
},
|
| 378 |
{
|
| 379 |
"epoch": 1.97,
|
| 380 |
+
"grad_norm": 0.08837890625,
|
| 381 |
+
"learning_rate": 0.00018333333333333334,
|
| 382 |
+
"loss": 0.4328,
|
| 383 |
"step": 336
|
| 384 |
},
|
| 385 |
{
|
| 386 |
"epoch": 2.02,
|
| 387 |
+
"grad_norm": 0.0751953125,
|
| 388 |
+
"learning_rate": 0.000175,
|
| 389 |
+
"loss": 0.43,
|
| 390 |
"step": 344
|
| 391 |
},
|
| 392 |
{
|
| 393 |
"epoch": 2.06,
|
| 394 |
+
"grad_norm": 0.087890625,
|
| 395 |
+
"learning_rate": 0.00016666666666666666,
|
| 396 |
+
"loss": 0.4169,
|
| 397 |
"step": 352
|
| 398 |
},
|
| 399 |
{
|
| 400 |
"epoch": 2.06,
|
| 401 |
+
"eval_loss": 0.3901583254337311,
|
| 402 |
+
"eval_runtime": 2.1777,
|
| 403 |
+
"eval_samples_per_second": 22.042,
|
| 404 |
+
"eval_steps_per_second": 2.755,
|
| 405 |
"step": 352
|
| 406 |
},
|
| 407 |
{
|
| 408 |
"epoch": 2.11,
|
| 409 |
+
"grad_norm": 0.0810546875,
|
| 410 |
+
"learning_rate": 0.00015833333333333332,
|
| 411 |
+
"loss": 0.3996,
|
| 412 |
"step": 360
|
| 413 |
},
|
| 414 |
{
|
| 415 |
"epoch": 2.16,
|
| 416 |
+
"grad_norm": 0.08154296875,
|
| 417 |
+
"learning_rate": 0.00015,
|
| 418 |
+
"loss": 0.4025,
|
| 419 |
"step": 368
|
| 420 |
},
|
| 421 |
{
|
| 422 |
"epoch": 2.21,
|
| 423 |
+
"grad_norm": 0.0791015625,
|
| 424 |
+
"learning_rate": 0.00014166666666666668,
|
| 425 |
+
"loss": 0.4098,
|
| 426 |
"step": 376
|
| 427 |
},
|
| 428 |
{
|
| 429 |
"epoch": 2.25,
|
| 430 |
+
"grad_norm": 0.0947265625,
|
| 431 |
+
"learning_rate": 0.00013333333333333334,
|
| 432 |
+
"loss": 0.4347,
|
| 433 |
"step": 384
|
| 434 |
},
|
| 435 |
{
|
| 436 |
"epoch": 2.25,
|
| 437 |
+
"eval_loss": 0.38746240735054016,
|
| 438 |
+
"eval_runtime": 2.1784,
|
| 439 |
+
"eval_samples_per_second": 22.035,
|
| 440 |
+
"eval_steps_per_second": 2.754,
|
| 441 |
"step": 384
|
| 442 |
},
|
| 443 |
{
|
| 444 |
"epoch": 2.3,
|
| 445 |
+
"grad_norm": 0.0859375,
|
| 446 |
+
"learning_rate": 0.000125,
|
| 447 |
+
"loss": 0.4234,
|
| 448 |
"step": 392
|
| 449 |
},
|
| 450 |
{
|
| 451 |
"epoch": 2.35,
|
| 452 |
+
"grad_norm": 0.08740234375,
|
| 453 |
+
"learning_rate": 0.00011666666666666667,
|
| 454 |
+
"loss": 0.4054,
|
| 455 |
"step": 400
|
| 456 |
},
|
| 457 |
{
|
| 458 |
"epoch": 2.39,
|
| 459 |
+
"grad_norm": 0.08984375,
|
| 460 |
+
"learning_rate": 0.00010833333333333334,
|
| 461 |
+
"loss": 0.4117,
|
| 462 |
"step": 408
|
| 463 |
},
|
| 464 |
{
|
| 465 |
"epoch": 2.44,
|
| 466 |
+
"grad_norm": 0.08935546875,
|
| 467 |
+
"learning_rate": 0.0001,
|
| 468 |
+
"loss": 0.4163,
|
| 469 |
"step": 416
|
| 470 |
},
|
| 471 |
{
|
| 472 |
"epoch": 2.44,
|
| 473 |
+
"eval_loss": 0.38652801513671875,
|
| 474 |
+
"eval_runtime": 2.1766,
|
| 475 |
+
"eval_samples_per_second": 22.053,
|
| 476 |
+
"eval_steps_per_second": 2.757,
|
| 477 |
"step": 416
|
| 478 |
},
|
| 479 |
{
|
| 480 |
"epoch": 2.49,
|
| 481 |
+
"grad_norm": 0.09619140625,
|
| 482 |
+
"learning_rate": 9.166666666666667e-05,
|
| 483 |
+
"loss": 0.4216,
|
| 484 |
"step": 424
|
| 485 |
},
|
| 486 |
{
|
| 487 |
"epoch": 2.53,
|
| 488 |
+
"grad_norm": 0.08544921875,
|
| 489 |
+
"learning_rate": 8.333333333333333e-05,
|
| 490 |
+
"loss": 0.3861,
|
| 491 |
"step": 432
|
| 492 |
},
|
| 493 |
{
|
| 494 |
"epoch": 2.58,
|
| 495 |
+
"grad_norm": 0.087890625,
|
| 496 |
+
"learning_rate": 7.5e-05,
|
| 497 |
+
"loss": 0.3994,
|
| 498 |
"step": 440
|
| 499 |
},
|
| 500 |
{
|
| 501 |
"epoch": 2.63,
|
| 502 |
+
"grad_norm": 0.08642578125,
|
| 503 |
+
"learning_rate": 6.666666666666667e-05,
|
| 504 |
+
"loss": 0.4171,
|
| 505 |
"step": 448
|
| 506 |
},
|
| 507 |
{
|
| 508 |
"epoch": 2.63,
|
| 509 |
+
"eval_loss": 0.38501664996147156,
|
| 510 |
+
"eval_runtime": 2.1793,
|
| 511 |
+
"eval_samples_per_second": 22.025,
|
| 512 |
+
"eval_steps_per_second": 2.753,
|
| 513 |
"step": 448
|
| 514 |
},
|
| 515 |
{
|
| 516 |
"epoch": 2.67,
|
| 517 |
+
"grad_norm": 0.0908203125,
|
| 518 |
+
"learning_rate": 5.833333333333333e-05,
|
| 519 |
+
"loss": 0.3924,
|
| 520 |
"step": 456
|
| 521 |
},
|
| 522 |
{
|
| 523 |
"epoch": 2.72,
|
| 524 |
+
"grad_norm": 0.09228515625,
|
| 525 |
+
"learning_rate": 5e-05,
|
| 526 |
+
"loss": 0.4003,
|
| 527 |
"step": 464
|
| 528 |
},
|
| 529 |
{
|
| 530 |
"epoch": 2.77,
|
| 531 |
+
"grad_norm": 0.08740234375,
|
| 532 |
+
"learning_rate": 4.1666666666666665e-05,
|
| 533 |
+
"loss": 0.4366,
|
| 534 |
"step": 472
|
| 535 |
},
|
| 536 |
{
|
| 537 |
"epoch": 2.82,
|
| 538 |
+
"grad_norm": 0.08544921875,
|
| 539 |
+
"learning_rate": 3.3333333333333335e-05,
|
| 540 |
+
"loss": 0.3856,
|
| 541 |
"step": 480
|
| 542 |
},
|
| 543 |
{
|
| 544 |
"epoch": 2.82,
|
| 545 |
+
"eval_loss": 0.3836025893688202,
|
| 546 |
+
"eval_runtime": 2.1811,
|
| 547 |
+
"eval_samples_per_second": 22.008,
|
| 548 |
+
"eval_steps_per_second": 2.751,
|
| 549 |
"step": 480
|
| 550 |
},
|
| 551 |
{
|
| 552 |
"epoch": 2.86,
|
| 553 |
+
"grad_norm": 0.0869140625,
|
| 554 |
+
"learning_rate": 2.5e-05,
|
| 555 |
+
"loss": 0.3991,
|
| 556 |
"step": 488
|
| 557 |
},
|
| 558 |
{
|
| 559 |
"epoch": 2.91,
|
| 560 |
+
"grad_norm": 0.09033203125,
|
| 561 |
+
"learning_rate": 1.6666666666666667e-05,
|
| 562 |
+
"loss": 0.402,
|
| 563 |
"step": 496
|
| 564 |
},
|
| 565 |
{
|
| 566 |
"epoch": 2.96,
|
| 567 |
+
"grad_norm": 0.08935546875,
|
| 568 |
+
"learning_rate": 8.333333333333334e-06,
|
| 569 |
+
"loss": 0.4184,
|
| 570 |
"step": 504
|
| 571 |
},
|
| 572 |
{
|
| 573 |
"epoch": 3.0,
|
| 574 |
+
"grad_norm": 0.0830078125,
|
| 575 |
"learning_rate": 0.0,
|
| 576 |
+
"loss": 0.3914,
|
| 577 |
"step": 512
|
| 578 |
},
|
| 579 |
{
|
| 580 |
"epoch": 3.0,
|
| 581 |
+
"eval_loss": 0.383393794298172,
|
| 582 |
+
"eval_runtime": 2.1833,
|
| 583 |
+
"eval_samples_per_second": 21.985,
|
| 584 |
+
"eval_steps_per_second": 2.748,
|
| 585 |
"step": 512
|
| 586 |
},
|
| 587 |
{
|
| 588 |
"epoch": 3.0,
|
| 589 |
"step": 512,
|
| 590 |
+
"total_flos": 9.863515762146509e+16,
|
| 591 |
+
"train_loss": 0.48598923115059733,
|
| 592 |
+
"train_runtime": 1129.8879,
|
| 593 |
+
"train_samples_per_second": 7.25,
|
| 594 |
+
"train_steps_per_second": 0.453
|
| 595 |
}
|
| 596 |
],
|
| 597 |
"logging_steps": 8,
|
|
|
|
| 599 |
"num_input_tokens_seen": 0,
|
| 600 |
"num_train_epochs": 4,
|
| 601 |
"save_steps": 32,
|
| 602 |
+
"total_flos": 9.863515762146509e+16,
|
| 603 |
"train_batch_size": 4,
|
| 604 |
"trial_name": null,
|
| 605 |
"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 4920
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8c822435a140bbc9ba485bc690b26c593635fea3e2667caf12b34f417b55a6ad
|
| 3 |
size 4920
|