Training checkpoint at step 4000
Browse files- trainer_state.json +1206 -126
trainer_state.json
CHANGED
|
@@ -1,373 +1,1453 @@
|
|
| 1 |
{
|
| 2 |
-
"best_global_step":
|
| 3 |
-
"best_metric": 2.
|
| 4 |
-
"best_model_checkpoint": "../runs/karpathy/fineweb-edu-100b-shuffle/meta-llama/Llama-3.2-1B/linear_adamw_wd1e-03_8x1024_mem32_bs64_hf_armt_dmem64/run_20/checkpoint-
|
| 5 |
-
"epoch": 0.
|
| 6 |
"eval_steps": 100,
|
| 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 |
"epoch": 0.0005,
|
| 14 |
-
"grad_norm":
|
| 15 |
"learning_rate": 4.8e-08,
|
| 16 |
-
"loss": 3.
|
| 17 |
"step": 25
|
| 18 |
},
|
| 19 |
{
|
| 20 |
"epoch": 0.001,
|
| 21 |
-
"grad_norm":
|
| 22 |
"learning_rate": 9.8e-08,
|
| 23 |
-
"loss": 3.
|
| 24 |
"step": 50
|
| 25 |
},
|
| 26 |
{
|
| 27 |
"epoch": 0.0015,
|
| 28 |
-
"grad_norm":
|
| 29 |
"learning_rate": 1.4800000000000003e-07,
|
| 30 |
-
"loss": 3.
|
| 31 |
"step": 75
|
| 32 |
},
|
| 33 |
{
|
| 34 |
"epoch": 0.002,
|
| 35 |
-
"grad_norm":
|
| 36 |
"learning_rate": 1.9800000000000003e-07,
|
| 37 |
-
"loss": 3.
|
| 38 |
"step": 100
|
| 39 |
},
|
| 40 |
{
|
| 41 |
"epoch": 0.002,
|
| 42 |
-
"eval_loss": 2.
|
| 43 |
-
"eval_runtime":
|
| 44 |
-
"eval_samples_per_second": 3.
|
| 45 |
-
"eval_steps_per_second": 1.
|
| 46 |
"step": 100
|
| 47 |
},
|
| 48 |
{
|
| 49 |
"epoch": 0.0025,
|
| 50 |
-
"grad_norm":
|
| 51 |
"learning_rate": 2.48e-07,
|
| 52 |
-
"loss": 2.
|
| 53 |
"step": 125
|
| 54 |
},
|
| 55 |
{
|
| 56 |
"epoch": 0.003,
|
| 57 |
-
"grad_norm":
|
| 58 |
"learning_rate": 2.9800000000000005e-07,
|
| 59 |
-
"loss": 2.
|
| 60 |
"step": 150
|
| 61 |
},
|
| 62 |
{
|
| 63 |
"epoch": 0.0035,
|
| 64 |
-
"grad_norm":
|
| 65 |
"learning_rate": 3.48e-07,
|
| 66 |
-
"loss": 2.
|
| 67 |
"step": 175
|
| 68 |
},
|
| 69 |
{
|
| 70 |
"epoch": 0.004,
|
| 71 |
-
"grad_norm": 1.
|
| 72 |
"learning_rate": 3.9800000000000004e-07,
|
| 73 |
-
"loss": 2.
|
| 74 |
"step": 200
|
| 75 |
},
|
| 76 |
{
|
| 77 |
"epoch": 0.004,
|
| 78 |
-
"eval_loss": 2.
|
| 79 |
-
"eval_runtime":
|
| 80 |
-
"eval_samples_per_second": 3.
|
| 81 |
-
"eval_steps_per_second": 1.
|
| 82 |
"step": 200
|
| 83 |
},
|
| 84 |
{
|
| 85 |
"epoch": 0.0045,
|
| 86 |
-
"grad_norm": 1.
|
| 87 |
"learning_rate": 4.4800000000000004e-07,
|
| 88 |
-
"loss": 2.
|
| 89 |
"step": 225
|
| 90 |
},
|
| 91 |
{
|
| 92 |
"epoch": 0.005,
|
| 93 |
-
"grad_norm": 1.
|
| 94 |
"learning_rate": 4.98e-07,
|
| 95 |
-
"loss": 2.
|
| 96 |
"step": 250
|
| 97 |
},
|
| 98 |
{
|
| 99 |
"epoch": 0.0055,
|
| 100 |
-
"grad_norm":
|
| 101 |
"learning_rate": 5.480000000000001e-07,
|
| 102 |
-
"loss": 2.
|
| 103 |
"step": 275
|
| 104 |
},
|
| 105 |
{
|
| 106 |
"epoch": 0.006,
|
| 107 |
-
"grad_norm":
|
| 108 |
"learning_rate": 5.98e-07,
|
| 109 |
-
"loss": 2.
|
| 110 |
"step": 300
|
| 111 |
},
|
| 112 |
{
|
| 113 |
"epoch": 0.006,
|
| 114 |
-
"eval_loss": 2.
|
| 115 |
-
"eval_runtime":
|
| 116 |
-
"eval_samples_per_second": 3.
|
| 117 |
-
"eval_steps_per_second": 1.
|
| 118 |
"step": 300
|
| 119 |
},
|
| 120 |
{
|
| 121 |
"epoch": 0.0065,
|
| 122 |
-
"grad_norm":
|
| 123 |
"learning_rate": 6.48e-07,
|
| 124 |
-
"loss": 2.
|
| 125 |
"step": 325
|
| 126 |
},
|
| 127 |
{
|
| 128 |
"epoch": 0.007,
|
| 129 |
-
"grad_norm": 0.
|
| 130 |
"learning_rate": 6.98e-07,
|
| 131 |
-
"loss": 2.
|
| 132 |
"step": 350
|
| 133 |
},
|
| 134 |
{
|
| 135 |
"epoch": 0.0075,
|
| 136 |
-
"grad_norm": 0.
|
| 137 |
"learning_rate": 7.480000000000001e-07,
|
| 138 |
-
"loss": 2.
|
| 139 |
"step": 375
|
| 140 |
},
|
| 141 |
{
|
| 142 |
"epoch": 0.008,
|
| 143 |
-
"grad_norm":
|
| 144 |
"learning_rate": 7.98e-07,
|
| 145 |
-
"loss": 2.
|
| 146 |
"step": 400
|
| 147 |
},
|
| 148 |
{
|
| 149 |
"epoch": 0.008,
|
| 150 |
-
"eval_loss": 2.
|
| 151 |
-
"eval_runtime":
|
| 152 |
-
"eval_samples_per_second": 3.
|
| 153 |
-
"eval_steps_per_second": 1.
|
| 154 |
"step": 400
|
| 155 |
},
|
| 156 |
{
|
| 157 |
"epoch": 0.0085,
|
| 158 |
-
"grad_norm":
|
| 159 |
"learning_rate": 8.480000000000001e-07,
|
| 160 |
-
"loss": 2.
|
| 161 |
"step": 425
|
| 162 |
},
|
| 163 |
{
|
| 164 |
"epoch": 0.009,
|
| 165 |
-
"grad_norm":
|
| 166 |
"learning_rate": 8.980000000000001e-07,
|
| 167 |
-
"loss": 2.
|
| 168 |
"step": 450
|
| 169 |
},
|
| 170 |
{
|
| 171 |
"epoch": 0.0095,
|
| 172 |
-
"grad_norm":
|
| 173 |
"learning_rate": 9.480000000000001e-07,
|
| 174 |
-
"loss": 2.
|
| 175 |
"step": 475
|
| 176 |
},
|
| 177 |
{
|
| 178 |
"epoch": 0.01,
|
| 179 |
-
"grad_norm":
|
| 180 |
"learning_rate": 9.98e-07,
|
| 181 |
-
"loss": 2.
|
| 182 |
"step": 500
|
| 183 |
},
|
| 184 |
{
|
| 185 |
"epoch": 0.01,
|
| 186 |
-
"eval_loss": 2.
|
| 187 |
-
"eval_runtime":
|
| 188 |
-
"eval_samples_per_second": 3.
|
| 189 |
-
"eval_steps_per_second": 1.
|
| 190 |
"step": 500
|
| 191 |
},
|
| 192 |
{
|
| 193 |
"epoch": 0.0105,
|
| 194 |
-
"grad_norm":
|
| 195 |
"learning_rate": 1.0480000000000002e-06,
|
| 196 |
-
"loss": 2.
|
| 197 |
"step": 525
|
| 198 |
},
|
| 199 |
{
|
| 200 |
"epoch": 0.011,
|
| 201 |
-
"grad_norm":
|
| 202 |
"learning_rate": 1.0980000000000001e-06,
|
| 203 |
-
"loss": 2.
|
| 204 |
"step": 550
|
| 205 |
},
|
| 206 |
{
|
| 207 |
"epoch": 0.0115,
|
| 208 |
-
"grad_norm":
|
| 209 |
"learning_rate": 1.148e-06,
|
| 210 |
-
"loss": 2.
|
| 211 |
"step": 575
|
| 212 |
},
|
| 213 |
{
|
| 214 |
"epoch": 0.012,
|
| 215 |
-
"grad_norm": 1.
|
| 216 |
"learning_rate": 1.1980000000000002e-06,
|
| 217 |
-
"loss": 2.
|
| 218 |
"step": 600
|
| 219 |
},
|
| 220 |
{
|
| 221 |
"epoch": 0.012,
|
| 222 |
-
"eval_loss": 2.
|
| 223 |
-
"eval_runtime":
|
| 224 |
-
"eval_samples_per_second": 3.
|
| 225 |
-
"eval_steps_per_second": 1.
|
| 226 |
"step": 600
|
| 227 |
},
|
| 228 |
{
|
| 229 |
"epoch": 0.0125,
|
| 230 |
-
"grad_norm":
|
| 231 |
"learning_rate": 1.248e-06,
|
| 232 |
-
"loss": 2.
|
| 233 |
"step": 625
|
| 234 |
},
|
| 235 |
{
|
| 236 |
"epoch": 0.013,
|
| 237 |
-
"grad_norm": 1.
|
| 238 |
"learning_rate": 1.2980000000000001e-06,
|
| 239 |
-
"loss": 2.
|
| 240 |
"step": 650
|
| 241 |
},
|
| 242 |
{
|
| 243 |
"epoch": 0.0135,
|
| 244 |
-
"grad_norm": 1.
|
| 245 |
"learning_rate": 1.348e-06,
|
| 246 |
-
"loss": 2.
|
| 247 |
"step": 675
|
| 248 |
},
|
| 249 |
{
|
| 250 |
"epoch": 0.014,
|
| 251 |
-
"grad_norm": 1.
|
| 252 |
"learning_rate": 1.3980000000000002e-06,
|
| 253 |
-
"loss": 2.
|
| 254 |
"step": 700
|
| 255 |
},
|
| 256 |
{
|
| 257 |
"epoch": 0.014,
|
| 258 |
-
"eval_loss": 2.
|
| 259 |
-
"eval_runtime":
|
| 260 |
-
"eval_samples_per_second": 3.
|
| 261 |
-
"eval_steps_per_second": 1.
|
| 262 |
"step": 700
|
| 263 |
},
|
| 264 |
{
|
| 265 |
"epoch": 0.0145,
|
| 266 |
-
"grad_norm": 1.
|
| 267 |
"learning_rate": 1.4480000000000002e-06,
|
| 268 |
-
"loss": 2.
|
| 269 |
"step": 725
|
| 270 |
},
|
| 271 |
{
|
| 272 |
"epoch": 0.015,
|
| 273 |
-
"grad_norm":
|
| 274 |
"learning_rate": 1.498e-06,
|
| 275 |
-
"loss": 2.
|
| 276 |
"step": 750
|
| 277 |
},
|
| 278 |
{
|
| 279 |
"epoch": 0.0155,
|
| 280 |
-
"grad_norm": 1.
|
| 281 |
"learning_rate": 1.548e-06,
|
| 282 |
-
"loss": 2.
|
| 283 |
"step": 775
|
| 284 |
},
|
| 285 |
{
|
| 286 |
"epoch": 0.016,
|
| 287 |
-
"grad_norm": 1.
|
| 288 |
"learning_rate": 1.5980000000000002e-06,
|
| 289 |
-
"loss": 2.
|
| 290 |
"step": 800
|
| 291 |
},
|
| 292 |
{
|
| 293 |
"epoch": 0.016,
|
| 294 |
-
"eval_loss": 2.
|
| 295 |
-
"eval_runtime":
|
| 296 |
-
"eval_samples_per_second":
|
| 297 |
-
"eval_steps_per_second": 1.
|
| 298 |
"step": 800
|
| 299 |
},
|
| 300 |
{
|
| 301 |
"epoch": 0.0165,
|
| 302 |
-
"grad_norm": 1.
|
| 303 |
"learning_rate": 1.6480000000000001e-06,
|
| 304 |
-
"loss": 2.
|
| 305 |
"step": 825
|
| 306 |
},
|
| 307 |
{
|
| 308 |
"epoch": 0.017,
|
| 309 |
-
"grad_norm": 1.
|
| 310 |
"learning_rate": 1.6980000000000003e-06,
|
| 311 |
-
"loss": 2.
|
| 312 |
"step": 850
|
| 313 |
},
|
| 314 |
{
|
| 315 |
"epoch": 0.0175,
|
| 316 |
-
"grad_norm": 1.
|
| 317 |
"learning_rate": 1.7480000000000002e-06,
|
| 318 |
-
"loss": 2.
|
| 319 |
"step": 875
|
| 320 |
},
|
| 321 |
{
|
| 322 |
"epoch": 0.018,
|
| 323 |
-
"grad_norm":
|
| 324 |
"learning_rate": 1.798e-06,
|
| 325 |
-
"loss": 2.
|
| 326 |
"step": 900
|
| 327 |
},
|
| 328 |
{
|
| 329 |
"epoch": 0.018,
|
| 330 |
-
"eval_loss": 2.
|
| 331 |
-
"eval_runtime":
|
| 332 |
-
"eval_samples_per_second": 3.
|
| 333 |
-
"eval_steps_per_second": 1.
|
| 334 |
"step": 900
|
| 335 |
},
|
| 336 |
{
|
| 337 |
"epoch": 0.0185,
|
| 338 |
-
"grad_norm": 1.
|
| 339 |
"learning_rate": 1.8480000000000001e-06,
|
| 340 |
-
"loss": 2.
|
| 341 |
"step": 925
|
| 342 |
},
|
| 343 |
{
|
| 344 |
"epoch": 0.019,
|
| 345 |
-
"grad_norm": 1.
|
| 346 |
"learning_rate": 1.898e-06,
|
| 347 |
-
"loss": 2.
|
| 348 |
"step": 950
|
| 349 |
},
|
| 350 |
{
|
| 351 |
"epoch": 0.0195,
|
| 352 |
-
"grad_norm": 1.
|
| 353 |
"learning_rate": 1.9480000000000002e-06,
|
| 354 |
-
"loss": 2.
|
| 355 |
"step": 975
|
| 356 |
},
|
| 357 |
{
|
| 358 |
"epoch": 0.02,
|
| 359 |
-
"grad_norm": 1.
|
| 360 |
"learning_rate": 1.998e-06,
|
| 361 |
-
"loss": 2.
|
| 362 |
"step": 1000
|
| 363 |
},
|
| 364 |
{
|
| 365 |
"epoch": 0.02,
|
| 366 |
-
"eval_loss": 2.
|
| 367 |
-
"eval_runtime":
|
| 368 |
-
"eval_samples_per_second":
|
| 369 |
-
"eval_steps_per_second": 1.
|
| 370 |
"step": 1000
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 371 |
}
|
| 372 |
],
|
| 373 |
"logging_steps": 25,
|
|
@@ -387,7 +1467,7 @@
|
|
| 387 |
"attributes": {}
|
| 388 |
}
|
| 389 |
},
|
| 390 |
-
"total_flos":
|
| 391 |
"train_batch_size": 1,
|
| 392 |
"trial_name": null,
|
| 393 |
"trial_params": null
|
|
|
|
| 1 |
{
|
| 2 |
+
"best_global_step": 4000,
|
| 3 |
+
"best_metric": 2.4323527812957764,
|
| 4 |
+
"best_model_checkpoint": "../runs/karpathy/fineweb-edu-100b-shuffle/meta-llama/Llama-3.2-1B/linear_adamw_wd1e-03_8x1024_mem32_bs64_hf_armt_dmem64/run_20/checkpoint-4000",
|
| 5 |
+
"epoch": 0.08,
|
| 6 |
"eval_steps": 100,
|
| 7 |
+
"global_step": 4000,
|
| 8 |
"is_hyper_param_search": false,
|
| 9 |
"is_local_process_zero": true,
|
| 10 |
"is_world_process_zero": true,
|
| 11 |
"log_history": [
|
| 12 |
{
|
| 13 |
"epoch": 0.0005,
|
| 14 |
+
"grad_norm": 39.75564521032967,
|
| 15 |
"learning_rate": 4.8e-08,
|
| 16 |
+
"loss": 3.6517,
|
| 17 |
"step": 25
|
| 18 |
},
|
| 19 |
{
|
| 20 |
"epoch": 0.001,
|
| 21 |
+
"grad_norm": 28.937531835097435,
|
| 22 |
"learning_rate": 9.8e-08,
|
| 23 |
+
"loss": 3.5931,
|
| 24 |
"step": 50
|
| 25 |
},
|
| 26 |
{
|
| 27 |
"epoch": 0.0015,
|
| 28 |
+
"grad_norm": 21.922720332659644,
|
| 29 |
"learning_rate": 1.4800000000000003e-07,
|
| 30 |
+
"loss": 3.3397,
|
| 31 |
"step": 75
|
| 32 |
},
|
| 33 |
{
|
| 34 |
"epoch": 0.002,
|
| 35 |
+
"grad_norm": 8.739610199908325,
|
| 36 |
"learning_rate": 1.9800000000000003e-07,
|
| 37 |
+
"loss": 3.1289,
|
| 38 |
"step": 100
|
| 39 |
},
|
| 40 |
{
|
| 41 |
"epoch": 0.002,
|
| 42 |
+
"eval_loss": 2.9243295192718506,
|
| 43 |
+
"eval_runtime": 264.3302,
|
| 44 |
+
"eval_samples_per_second": 3.11,
|
| 45 |
+
"eval_steps_per_second": 1.555,
|
| 46 |
"step": 100
|
| 47 |
},
|
| 48 |
{
|
| 49 |
"epoch": 0.0025,
|
| 50 |
+
"grad_norm": 4.433912600039677,
|
| 51 |
"learning_rate": 2.48e-07,
|
| 52 |
+
"loss": 2.8957,
|
| 53 |
"step": 125
|
| 54 |
},
|
| 55 |
{
|
| 56 |
"epoch": 0.003,
|
| 57 |
+
"grad_norm": 3.2874790066620303,
|
| 58 |
"learning_rate": 2.9800000000000005e-07,
|
| 59 |
+
"loss": 2.763,
|
| 60 |
"step": 150
|
| 61 |
},
|
| 62 |
{
|
| 63 |
"epoch": 0.0035,
|
| 64 |
+
"grad_norm": 1.5203472215469231,
|
| 65 |
"learning_rate": 3.48e-07,
|
| 66 |
+
"loss": 2.676,
|
| 67 |
"step": 175
|
| 68 |
},
|
| 69 |
{
|
| 70 |
"epoch": 0.004,
|
| 71 |
+
"grad_norm": 1.1945541683905954,
|
| 72 |
"learning_rate": 3.9800000000000004e-07,
|
| 73 |
+
"loss": 2.635,
|
| 74 |
"step": 200
|
| 75 |
},
|
| 76 |
{
|
| 77 |
"epoch": 0.004,
|
| 78 |
+
"eval_loss": 2.6094932556152344,
|
| 79 |
+
"eval_runtime": 265.7702,
|
| 80 |
+
"eval_samples_per_second": 3.093,
|
| 81 |
+
"eval_steps_per_second": 1.546,
|
| 82 |
"step": 200
|
| 83 |
},
|
| 84 |
{
|
| 85 |
"epoch": 0.0045,
|
| 86 |
+
"grad_norm": 1.0852713304633745,
|
| 87 |
"learning_rate": 4.4800000000000004e-07,
|
| 88 |
+
"loss": 2.6016,
|
| 89 |
"step": 225
|
| 90 |
},
|
| 91 |
{
|
| 92 |
"epoch": 0.005,
|
| 93 |
+
"grad_norm": 1.0733940346699529,
|
| 94 |
"learning_rate": 4.98e-07,
|
| 95 |
+
"loss": 2.5797,
|
| 96 |
"step": 250
|
| 97 |
},
|
| 98 |
{
|
| 99 |
"epoch": 0.0055,
|
| 100 |
+
"grad_norm": 0.9273949035031271,
|
| 101 |
"learning_rate": 5.480000000000001e-07,
|
| 102 |
+
"loss": 2.5607,
|
| 103 |
"step": 275
|
| 104 |
},
|
| 105 |
{
|
| 106 |
"epoch": 0.006,
|
| 107 |
+
"grad_norm": 0.9289300678591714,
|
| 108 |
"learning_rate": 5.98e-07,
|
| 109 |
+
"loss": 2.552,
|
| 110 |
"step": 300
|
| 111 |
},
|
| 112 |
{
|
| 113 |
"epoch": 0.006,
|
| 114 |
+
"eval_loss": 2.541522264480591,
|
| 115 |
+
"eval_runtime": 266.7478,
|
| 116 |
+
"eval_samples_per_second": 3.082,
|
| 117 |
+
"eval_steps_per_second": 1.541,
|
| 118 |
"step": 300
|
| 119 |
},
|
| 120 |
{
|
| 121 |
"epoch": 0.0065,
|
| 122 |
+
"grad_norm": 1.1328584507449984,
|
| 123 |
"learning_rate": 6.48e-07,
|
| 124 |
+
"loss": 2.5402,
|
| 125 |
"step": 325
|
| 126 |
},
|
| 127 |
{
|
| 128 |
"epoch": 0.007,
|
| 129 |
+
"grad_norm": 0.8593307029257858,
|
| 130 |
"learning_rate": 6.98e-07,
|
| 131 |
+
"loss": 2.5286,
|
| 132 |
"step": 350
|
| 133 |
},
|
| 134 |
{
|
| 135 |
"epoch": 0.0075,
|
| 136 |
+
"grad_norm": 0.895615604067586,
|
| 137 |
"learning_rate": 7.480000000000001e-07,
|
| 138 |
+
"loss": 2.5311,
|
| 139 |
"step": 375
|
| 140 |
},
|
| 141 |
{
|
| 142 |
"epoch": 0.008,
|
| 143 |
+
"grad_norm": 0.912306580242149,
|
| 144 |
"learning_rate": 7.98e-07,
|
| 145 |
+
"loss": 2.5037,
|
| 146 |
"step": 400
|
| 147 |
},
|
| 148 |
{
|
| 149 |
"epoch": 0.008,
|
| 150 |
+
"eval_loss": 2.514389991760254,
|
| 151 |
+
"eval_runtime": 266.4899,
|
| 152 |
+
"eval_samples_per_second": 3.085,
|
| 153 |
+
"eval_steps_per_second": 1.542,
|
| 154 |
"step": 400
|
| 155 |
},
|
| 156 |
{
|
| 157 |
"epoch": 0.0085,
|
| 158 |
+
"grad_norm": 1.1866535514670034,
|
| 159 |
"learning_rate": 8.480000000000001e-07,
|
| 160 |
+
"loss": 2.5011,
|
| 161 |
"step": 425
|
| 162 |
},
|
| 163 |
{
|
| 164 |
"epoch": 0.009,
|
| 165 |
+
"grad_norm": 1.211342504193914,
|
| 166 |
"learning_rate": 8.980000000000001e-07,
|
| 167 |
+
"loss": 2.503,
|
| 168 |
"step": 450
|
| 169 |
},
|
| 170 |
{
|
| 171 |
"epoch": 0.0095,
|
| 172 |
+
"grad_norm": 1.113763817383069,
|
| 173 |
"learning_rate": 9.480000000000001e-07,
|
| 174 |
+
"loss": 2.4999,
|
| 175 |
"step": 475
|
| 176 |
},
|
| 177 |
{
|
| 178 |
"epoch": 0.01,
|
| 179 |
+
"grad_norm": 1.2585585589647226,
|
| 180 |
"learning_rate": 9.98e-07,
|
| 181 |
+
"loss": 2.4872,
|
| 182 |
"step": 500
|
| 183 |
},
|
| 184 |
{
|
| 185 |
"epoch": 0.01,
|
| 186 |
+
"eval_loss": 2.497868061065674,
|
| 187 |
+
"eval_runtime": 265.7962,
|
| 188 |
+
"eval_samples_per_second": 3.093,
|
| 189 |
+
"eval_steps_per_second": 1.546,
|
| 190 |
"step": 500
|
| 191 |
},
|
| 192 |
{
|
| 193 |
"epoch": 0.0105,
|
| 194 |
+
"grad_norm": 1.2585825718084245,
|
| 195 |
"learning_rate": 1.0480000000000002e-06,
|
| 196 |
+
"loss": 2.4852,
|
| 197 |
"step": 525
|
| 198 |
},
|
| 199 |
{
|
| 200 |
"epoch": 0.011,
|
| 201 |
+
"grad_norm": 1.4101257437846046,
|
| 202 |
"learning_rate": 1.0980000000000001e-06,
|
| 203 |
+
"loss": 2.4892,
|
| 204 |
"step": 550
|
| 205 |
},
|
| 206 |
{
|
| 207 |
"epoch": 0.0115,
|
| 208 |
+
"grad_norm": 1.1975234150707363,
|
| 209 |
"learning_rate": 1.148e-06,
|
| 210 |
+
"loss": 2.4861,
|
| 211 |
"step": 575
|
| 212 |
},
|
| 213 |
{
|
| 214 |
"epoch": 0.012,
|
| 215 |
+
"grad_norm": 1.3662769225582332,
|
| 216 |
"learning_rate": 1.1980000000000002e-06,
|
| 217 |
+
"loss": 2.4882,
|
| 218 |
"step": 600
|
| 219 |
},
|
| 220 |
{
|
| 221 |
"epoch": 0.012,
|
| 222 |
+
"eval_loss": 2.4879231452941895,
|
| 223 |
+
"eval_runtime": 267.0005,
|
| 224 |
+
"eval_samples_per_second": 3.079,
|
| 225 |
+
"eval_steps_per_second": 1.539,
|
| 226 |
"step": 600
|
| 227 |
},
|
| 228 |
{
|
| 229 |
"epoch": 0.0125,
|
| 230 |
+
"grad_norm": 1.3086724275194024,
|
| 231 |
"learning_rate": 1.248e-06,
|
| 232 |
+
"loss": 2.4745,
|
| 233 |
"step": 625
|
| 234 |
},
|
| 235 |
{
|
| 236 |
"epoch": 0.013,
|
| 237 |
+
"grad_norm": 1.317023206802888,
|
| 238 |
"learning_rate": 1.2980000000000001e-06,
|
| 239 |
+
"loss": 2.4727,
|
| 240 |
"step": 650
|
| 241 |
},
|
| 242 |
{
|
| 243 |
"epoch": 0.0135,
|
| 244 |
+
"grad_norm": 1.5284967544483212,
|
| 245 |
"learning_rate": 1.348e-06,
|
| 246 |
+
"loss": 2.469,
|
| 247 |
"step": 675
|
| 248 |
},
|
| 249 |
{
|
| 250 |
"epoch": 0.014,
|
| 251 |
+
"grad_norm": 1.1047595217316941,
|
| 252 |
"learning_rate": 1.3980000000000002e-06,
|
| 253 |
+
"loss": 2.4695,
|
| 254 |
"step": 700
|
| 255 |
},
|
| 256 |
{
|
| 257 |
"epoch": 0.014,
|
| 258 |
+
"eval_loss": 2.480103015899658,
|
| 259 |
+
"eval_runtime": 263.5022,
|
| 260 |
+
"eval_samples_per_second": 3.12,
|
| 261 |
+
"eval_steps_per_second": 1.56,
|
| 262 |
"step": 700
|
| 263 |
},
|
| 264 |
{
|
| 265 |
"epoch": 0.0145,
|
| 266 |
+
"grad_norm": 1.2077328209863791,
|
| 267 |
"learning_rate": 1.4480000000000002e-06,
|
| 268 |
+
"loss": 2.4654,
|
| 269 |
"step": 725
|
| 270 |
},
|
| 271 |
{
|
| 272 |
"epoch": 0.015,
|
| 273 |
+
"grad_norm": 1.209220841771836,
|
| 274 |
"learning_rate": 1.498e-06,
|
| 275 |
+
"loss": 2.4663,
|
| 276 |
"step": 750
|
| 277 |
},
|
| 278 |
{
|
| 279 |
"epoch": 0.0155,
|
| 280 |
+
"grad_norm": 1.3063169829879686,
|
| 281 |
"learning_rate": 1.548e-06,
|
| 282 |
+
"loss": 2.4704,
|
| 283 |
"step": 775
|
| 284 |
},
|
| 285 |
{
|
| 286 |
"epoch": 0.016,
|
| 287 |
+
"grad_norm": 1.3180183352683195,
|
| 288 |
"learning_rate": 1.5980000000000002e-06,
|
| 289 |
+
"loss": 2.4583,
|
| 290 |
"step": 800
|
| 291 |
},
|
| 292 |
{
|
| 293 |
"epoch": 0.016,
|
| 294 |
+
"eval_loss": 2.473590850830078,
|
| 295 |
+
"eval_runtime": 305.9875,
|
| 296 |
+
"eval_samples_per_second": 2.686,
|
| 297 |
+
"eval_steps_per_second": 1.343,
|
| 298 |
"step": 800
|
| 299 |
},
|
| 300 |
{
|
| 301 |
"epoch": 0.0165,
|
| 302 |
+
"grad_norm": 1.1674852380778837,
|
| 303 |
"learning_rate": 1.6480000000000001e-06,
|
| 304 |
+
"loss": 2.467,
|
| 305 |
"step": 825
|
| 306 |
},
|
| 307 |
{
|
| 308 |
"epoch": 0.017,
|
| 309 |
+
"grad_norm": 1.2497656349941002,
|
| 310 |
"learning_rate": 1.6980000000000003e-06,
|
| 311 |
+
"loss": 2.4612,
|
| 312 |
"step": 850
|
| 313 |
},
|
| 314 |
{
|
| 315 |
"epoch": 0.0175,
|
| 316 |
+
"grad_norm": 1.3358614980967494,
|
| 317 |
"learning_rate": 1.7480000000000002e-06,
|
| 318 |
+
"loss": 2.4636,
|
| 319 |
"step": 875
|
| 320 |
},
|
| 321 |
{
|
| 322 |
"epoch": 0.018,
|
| 323 |
+
"grad_norm": 1.252489857653356,
|
| 324 |
"learning_rate": 1.798e-06,
|
| 325 |
+
"loss": 2.454,
|
| 326 |
"step": 900
|
| 327 |
},
|
| 328 |
{
|
| 329 |
"epoch": 0.018,
|
| 330 |
+
"eval_loss": 2.4681763648986816,
|
| 331 |
+
"eval_runtime": 264.702,
|
| 332 |
+
"eval_samples_per_second": 3.105,
|
| 333 |
+
"eval_steps_per_second": 1.553,
|
| 334 |
"step": 900
|
| 335 |
},
|
| 336 |
{
|
| 337 |
"epoch": 0.0185,
|
| 338 |
+
"grad_norm": 1.2815437998994337,
|
| 339 |
"learning_rate": 1.8480000000000001e-06,
|
| 340 |
+
"loss": 2.4571,
|
| 341 |
"step": 925
|
| 342 |
},
|
| 343 |
{
|
| 344 |
"epoch": 0.019,
|
| 345 |
+
"grad_norm": 1.0902475329451575,
|
| 346 |
"learning_rate": 1.898e-06,
|
| 347 |
+
"loss": 2.451,
|
| 348 |
"step": 950
|
| 349 |
},
|
| 350 |
{
|
| 351 |
"epoch": 0.0195,
|
| 352 |
+
"grad_norm": 1.1502696024965324,
|
| 353 |
"learning_rate": 1.9480000000000002e-06,
|
| 354 |
+
"loss": 2.4527,
|
| 355 |
"step": 975
|
| 356 |
},
|
| 357 |
{
|
| 358 |
"epoch": 0.02,
|
| 359 |
+
"grad_norm": 1.2336661855806117,
|
| 360 |
"learning_rate": 1.998e-06,
|
| 361 |
+
"loss": 2.4496,
|
| 362 |
"step": 1000
|
| 363 |
},
|
| 364 |
{
|
| 365 |
"epoch": 0.02,
|
| 366 |
+
"eval_loss": 2.463880777359009,
|
| 367 |
+
"eval_runtime": 275.7426,
|
| 368 |
+
"eval_samples_per_second": 2.981,
|
| 369 |
+
"eval_steps_per_second": 1.491,
|
| 370 |
"step": 1000
|
| 371 |
+
},
|
| 372 |
+
{
|
| 373 |
+
"epoch": 0.0205,
|
| 374 |
+
"grad_norm": 1.2680742209094296,
|
| 375 |
+
"learning_rate": 2.048e-06,
|
| 376 |
+
"loss": 2.4494,
|
| 377 |
+
"step": 1025
|
| 378 |
+
},
|
| 379 |
+
{
|
| 380 |
+
"epoch": 0.021,
|
| 381 |
+
"grad_norm": 1.0341778808278126,
|
| 382 |
+
"learning_rate": 2.098e-06,
|
| 383 |
+
"loss": 2.4467,
|
| 384 |
+
"step": 1050
|
| 385 |
+
},
|
| 386 |
+
{
|
| 387 |
+
"epoch": 0.0215,
|
| 388 |
+
"grad_norm": 0.9860490736001175,
|
| 389 |
+
"learning_rate": 2.148e-06,
|
| 390 |
+
"loss": 2.4473,
|
| 391 |
+
"step": 1075
|
| 392 |
+
},
|
| 393 |
+
{
|
| 394 |
+
"epoch": 0.022,
|
| 395 |
+
"grad_norm": 0.9419267295275278,
|
| 396 |
+
"learning_rate": 2.198e-06,
|
| 397 |
+
"loss": 2.443,
|
| 398 |
+
"step": 1100
|
| 399 |
+
},
|
| 400 |
+
{
|
| 401 |
+
"epoch": 0.022,
|
| 402 |
+
"eval_loss": 2.4598941802978516,
|
| 403 |
+
"eval_runtime": 265.0502,
|
| 404 |
+
"eval_samples_per_second": 3.101,
|
| 405 |
+
"eval_steps_per_second": 1.551,
|
| 406 |
+
"step": 1100
|
| 407 |
+
},
|
| 408 |
+
{
|
| 409 |
+
"epoch": 0.0225,
|
| 410 |
+
"grad_norm": 1.3280720471027394,
|
| 411 |
+
"learning_rate": 2.2480000000000003e-06,
|
| 412 |
+
"loss": 2.4515,
|
| 413 |
+
"step": 1125
|
| 414 |
+
},
|
| 415 |
+
{
|
| 416 |
+
"epoch": 0.023,
|
| 417 |
+
"grad_norm": 1.053570785582915,
|
| 418 |
+
"learning_rate": 2.2980000000000003e-06,
|
| 419 |
+
"loss": 2.4396,
|
| 420 |
+
"step": 1150
|
| 421 |
+
},
|
| 422 |
+
{
|
| 423 |
+
"epoch": 0.0235,
|
| 424 |
+
"grad_norm": 0.9108119839585552,
|
| 425 |
+
"learning_rate": 2.3480000000000002e-06,
|
| 426 |
+
"loss": 2.4442,
|
| 427 |
+
"step": 1175
|
| 428 |
+
},
|
| 429 |
+
{
|
| 430 |
+
"epoch": 0.024,
|
| 431 |
+
"grad_norm": 1.0062346367900277,
|
| 432 |
+
"learning_rate": 2.398e-06,
|
| 433 |
+
"loss": 2.4443,
|
| 434 |
+
"step": 1200
|
| 435 |
+
},
|
| 436 |
+
{
|
| 437 |
+
"epoch": 0.024,
|
| 438 |
+
"eval_loss": 2.456455945968628,
|
| 439 |
+
"eval_runtime": 264.5888,
|
| 440 |
+
"eval_samples_per_second": 3.107,
|
| 441 |
+
"eval_steps_per_second": 1.553,
|
| 442 |
+
"step": 1200
|
| 443 |
+
},
|
| 444 |
+
{
|
| 445 |
+
"epoch": 0.0245,
|
| 446 |
+
"grad_norm": 1.0264127705426926,
|
| 447 |
+
"learning_rate": 2.448e-06,
|
| 448 |
+
"loss": 2.4351,
|
| 449 |
+
"step": 1225
|
| 450 |
+
},
|
| 451 |
+
{
|
| 452 |
+
"epoch": 0.025,
|
| 453 |
+
"grad_norm": 0.8015249588347212,
|
| 454 |
+
"learning_rate": 2.498e-06,
|
| 455 |
+
"loss": 2.4406,
|
| 456 |
+
"step": 1250
|
| 457 |
+
},
|
| 458 |
+
{
|
| 459 |
+
"epoch": 0.0255,
|
| 460 |
+
"grad_norm": 1.1105649485540114,
|
| 461 |
+
"learning_rate": 2.5480000000000004e-06,
|
| 462 |
+
"loss": 2.4377,
|
| 463 |
+
"step": 1275
|
| 464 |
+
},
|
| 465 |
+
{
|
| 466 |
+
"epoch": 0.026,
|
| 467 |
+
"grad_norm": 0.9701758426012801,
|
| 468 |
+
"learning_rate": 2.598e-06,
|
| 469 |
+
"loss": 2.4341,
|
| 470 |
+
"step": 1300
|
| 471 |
+
},
|
| 472 |
+
{
|
| 473 |
+
"epoch": 0.026,
|
| 474 |
+
"eval_loss": 2.453026056289673,
|
| 475 |
+
"eval_runtime": 264.7653,
|
| 476 |
+
"eval_samples_per_second": 3.105,
|
| 477 |
+
"eval_steps_per_second": 1.552,
|
| 478 |
+
"step": 1300
|
| 479 |
+
},
|
| 480 |
+
{
|
| 481 |
+
"epoch": 0.0265,
|
| 482 |
+
"grad_norm": 0.9587254891845429,
|
| 483 |
+
"learning_rate": 2.648e-06,
|
| 484 |
+
"loss": 2.4303,
|
| 485 |
+
"step": 1325
|
| 486 |
+
},
|
| 487 |
+
{
|
| 488 |
+
"epoch": 0.027,
|
| 489 |
+
"grad_norm": 0.8135883960763247,
|
| 490 |
+
"learning_rate": 2.6980000000000003e-06,
|
| 491 |
+
"loss": 2.4363,
|
| 492 |
+
"step": 1350
|
| 493 |
+
},
|
| 494 |
+
{
|
| 495 |
+
"epoch": 0.0275,
|
| 496 |
+
"grad_norm": 0.9192860127847176,
|
| 497 |
+
"learning_rate": 2.748e-06,
|
| 498 |
+
"loss": 2.4257,
|
| 499 |
+
"step": 1375
|
| 500 |
+
},
|
| 501 |
+
{
|
| 502 |
+
"epoch": 0.028,
|
| 503 |
+
"grad_norm": 0.947465928893444,
|
| 504 |
+
"learning_rate": 2.798e-06,
|
| 505 |
+
"loss": 2.4353,
|
| 506 |
+
"step": 1400
|
| 507 |
+
},
|
| 508 |
+
{
|
| 509 |
+
"epoch": 0.028,
|
| 510 |
+
"eval_loss": 2.450345993041992,
|
| 511 |
+
"eval_runtime": 265.6266,
|
| 512 |
+
"eval_samples_per_second": 3.095,
|
| 513 |
+
"eval_steps_per_second": 1.547,
|
| 514 |
+
"step": 1400
|
| 515 |
+
},
|
| 516 |
+
{
|
| 517 |
+
"epoch": 0.0285,
|
| 518 |
+
"grad_norm": 0.9270137901066681,
|
| 519 |
+
"learning_rate": 2.848e-06,
|
| 520 |
+
"loss": 2.4347,
|
| 521 |
+
"step": 1425
|
| 522 |
+
},
|
| 523 |
+
{
|
| 524 |
+
"epoch": 0.029,
|
| 525 |
+
"grad_norm": 0.8839980710491563,
|
| 526 |
+
"learning_rate": 2.8980000000000005e-06,
|
| 527 |
+
"loss": 2.4213,
|
| 528 |
+
"step": 1450
|
| 529 |
+
},
|
| 530 |
+
{
|
| 531 |
+
"epoch": 0.0295,
|
| 532 |
+
"grad_norm": 0.913196005454606,
|
| 533 |
+
"learning_rate": 2.9480000000000004e-06,
|
| 534 |
+
"loss": 2.4232,
|
| 535 |
+
"step": 1475
|
| 536 |
+
},
|
| 537 |
+
{
|
| 538 |
+
"epoch": 0.03,
|
| 539 |
+
"grad_norm": 0.8139623858623861,
|
| 540 |
+
"learning_rate": 2.9980000000000003e-06,
|
| 541 |
+
"loss": 2.4254,
|
| 542 |
+
"step": 1500
|
| 543 |
+
},
|
| 544 |
+
{
|
| 545 |
+
"epoch": 0.03,
|
| 546 |
+
"eval_loss": 2.447662830352783,
|
| 547 |
+
"eval_runtime": 263.4353,
|
| 548 |
+
"eval_samples_per_second": 3.12,
|
| 549 |
+
"eval_steps_per_second": 1.56,
|
| 550 |
+
"step": 1500
|
| 551 |
+
},
|
| 552 |
+
{
|
| 553 |
+
"epoch": 0.0305,
|
| 554 |
+
"grad_norm": 0.8422198221554755,
|
| 555 |
+
"learning_rate": 3.0480000000000003e-06,
|
| 556 |
+
"loss": 2.4196,
|
| 557 |
+
"step": 1525
|
| 558 |
+
},
|
| 559 |
+
{
|
| 560 |
+
"epoch": 0.031,
|
| 561 |
+
"grad_norm": 0.8542957579365906,
|
| 562 |
+
"learning_rate": 3.0980000000000007e-06,
|
| 563 |
+
"loss": 2.4294,
|
| 564 |
+
"step": 1550
|
| 565 |
+
},
|
| 566 |
+
{
|
| 567 |
+
"epoch": 0.0315,
|
| 568 |
+
"grad_norm": 1.149263137594797,
|
| 569 |
+
"learning_rate": 3.1480000000000006e-06,
|
| 570 |
+
"loss": 2.4265,
|
| 571 |
+
"step": 1575
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"epoch": 0.032,
|
| 575 |
+
"grad_norm": 0.811470126240392,
|
| 576 |
+
"learning_rate": 3.198e-06,
|
| 577 |
+
"loss": 2.4105,
|
| 578 |
+
"step": 1600
|
| 579 |
+
},
|
| 580 |
+
{
|
| 581 |
+
"epoch": 0.032,
|
| 582 |
+
"eval_loss": 2.4456679821014404,
|
| 583 |
+
"eval_runtime": 264.056,
|
| 584 |
+
"eval_samples_per_second": 3.113,
|
| 585 |
+
"eval_steps_per_second": 1.556,
|
| 586 |
+
"step": 1600
|
| 587 |
+
},
|
| 588 |
+
{
|
| 589 |
+
"epoch": 0.0325,
|
| 590 |
+
"grad_norm": 2.3928975221881434,
|
| 591 |
+
"learning_rate": 3.248e-06,
|
| 592 |
+
"loss": 2.4208,
|
| 593 |
+
"step": 1625
|
| 594 |
+
},
|
| 595 |
+
{
|
| 596 |
+
"epoch": 0.033,
|
| 597 |
+
"grad_norm": 0.8031315125360012,
|
| 598 |
+
"learning_rate": 3.298e-06,
|
| 599 |
+
"loss": 2.4224,
|
| 600 |
+
"step": 1650
|
| 601 |
+
},
|
| 602 |
+
{
|
| 603 |
+
"epoch": 0.0335,
|
| 604 |
+
"grad_norm": 0.835567276692195,
|
| 605 |
+
"learning_rate": 3.348e-06,
|
| 606 |
+
"loss": 2.4188,
|
| 607 |
+
"step": 1675
|
| 608 |
+
},
|
| 609 |
+
{
|
| 610 |
+
"epoch": 0.034,
|
| 611 |
+
"grad_norm": 0.8894325175719718,
|
| 612 |
+
"learning_rate": 3.3980000000000003e-06,
|
| 613 |
+
"loss": 2.4206,
|
| 614 |
+
"step": 1700
|
| 615 |
+
},
|
| 616 |
+
{
|
| 617 |
+
"epoch": 0.034,
|
| 618 |
+
"eval_loss": 2.4437851905822754,
|
| 619 |
+
"eval_runtime": 264.6455,
|
| 620 |
+
"eval_samples_per_second": 3.106,
|
| 621 |
+
"eval_steps_per_second": 1.553,
|
| 622 |
+
"step": 1700
|
| 623 |
+
},
|
| 624 |
+
{
|
| 625 |
+
"epoch": 0.0345,
|
| 626 |
+
"grad_norm": 0.802724390649243,
|
| 627 |
+
"learning_rate": 3.4480000000000003e-06,
|
| 628 |
+
"loss": 2.4241,
|
| 629 |
+
"step": 1725
|
| 630 |
+
},
|
| 631 |
+
{
|
| 632 |
+
"epoch": 0.035,
|
| 633 |
+
"grad_norm": 0.8206312612014312,
|
| 634 |
+
"learning_rate": 3.4980000000000002e-06,
|
| 635 |
+
"loss": 2.4157,
|
| 636 |
+
"step": 1750
|
| 637 |
+
},
|
| 638 |
+
{
|
| 639 |
+
"epoch": 0.0355,
|
| 640 |
+
"grad_norm": 0.8653789917535344,
|
| 641 |
+
"learning_rate": 3.548e-06,
|
| 642 |
+
"loss": 2.412,
|
| 643 |
+
"step": 1775
|
| 644 |
+
},
|
| 645 |
+
{
|
| 646 |
+
"epoch": 0.036,
|
| 647 |
+
"grad_norm": 0.7816319078215015,
|
| 648 |
+
"learning_rate": 3.5980000000000005e-06,
|
| 649 |
+
"loss": 2.4179,
|
| 650 |
+
"step": 1800
|
| 651 |
+
},
|
| 652 |
+
{
|
| 653 |
+
"epoch": 0.036,
|
| 654 |
+
"eval_loss": 2.4423036575317383,
|
| 655 |
+
"eval_runtime": 264.5578,
|
| 656 |
+
"eval_samples_per_second": 3.107,
|
| 657 |
+
"eval_steps_per_second": 1.554,
|
| 658 |
+
"step": 1800
|
| 659 |
+
},
|
| 660 |
+
{
|
| 661 |
+
"epoch": 0.0365,
|
| 662 |
+
"grad_norm": 0.707594544466941,
|
| 663 |
+
"learning_rate": 3.6480000000000005e-06,
|
| 664 |
+
"loss": 2.416,
|
| 665 |
+
"step": 1825
|
| 666 |
+
},
|
| 667 |
+
{
|
| 668 |
+
"epoch": 0.037,
|
| 669 |
+
"grad_norm": 0.7481066913011816,
|
| 670 |
+
"learning_rate": 3.6980000000000004e-06,
|
| 671 |
+
"loss": 2.4242,
|
| 672 |
+
"step": 1850
|
| 673 |
+
},
|
| 674 |
+
{
|
| 675 |
+
"epoch": 0.0375,
|
| 676 |
+
"grad_norm": 0.7612014979445353,
|
| 677 |
+
"learning_rate": 3.7480000000000004e-06,
|
| 678 |
+
"loss": 2.4173,
|
| 679 |
+
"step": 1875
|
| 680 |
+
},
|
| 681 |
+
{
|
| 682 |
+
"epoch": 0.038,
|
| 683 |
+
"grad_norm": 0.772750918048857,
|
| 684 |
+
"learning_rate": 3.7980000000000007e-06,
|
| 685 |
+
"loss": 2.4134,
|
| 686 |
+
"step": 1900
|
| 687 |
+
},
|
| 688 |
+
{
|
| 689 |
+
"epoch": 0.038,
|
| 690 |
+
"eval_loss": 2.440969228744507,
|
| 691 |
+
"eval_runtime": 274.3624,
|
| 692 |
+
"eval_samples_per_second": 2.996,
|
| 693 |
+
"eval_steps_per_second": 1.498,
|
| 694 |
+
"step": 1900
|
| 695 |
+
},
|
| 696 |
+
{
|
| 697 |
+
"epoch": 0.0385,
|
| 698 |
+
"grad_norm": 0.7927966042188935,
|
| 699 |
+
"learning_rate": 3.848e-06,
|
| 700 |
+
"loss": 2.4131,
|
| 701 |
+
"step": 1925
|
| 702 |
+
},
|
| 703 |
+
{
|
| 704 |
+
"epoch": 0.039,
|
| 705 |
+
"grad_norm": 0.7664274167276341,
|
| 706 |
+
"learning_rate": 3.898e-06,
|
| 707 |
+
"loss": 2.4133,
|
| 708 |
+
"step": 1950
|
| 709 |
+
},
|
| 710 |
+
{
|
| 711 |
+
"epoch": 0.0395,
|
| 712 |
+
"grad_norm": 0.7038638213491795,
|
| 713 |
+
"learning_rate": 3.948e-06,
|
| 714 |
+
"loss": 2.4135,
|
| 715 |
+
"step": 1975
|
| 716 |
+
},
|
| 717 |
+
{
|
| 718 |
+
"epoch": 0.04,
|
| 719 |
+
"grad_norm": 0.7231696877425319,
|
| 720 |
+
"learning_rate": 3.9980000000000005e-06,
|
| 721 |
+
"loss": 2.4169,
|
| 722 |
+
"step": 2000
|
| 723 |
+
},
|
| 724 |
+
{
|
| 725 |
+
"epoch": 0.04,
|
| 726 |
+
"eval_loss": 2.439641237258911,
|
| 727 |
+
"eval_runtime": 282.4449,
|
| 728 |
+
"eval_samples_per_second": 2.91,
|
| 729 |
+
"eval_steps_per_second": 1.455,
|
| 730 |
+
"step": 2000
|
| 731 |
+
},
|
| 732 |
+
{
|
| 733 |
+
"epoch": 0.0405,
|
| 734 |
+
"grad_norm": 0.7184393791203537,
|
| 735 |
+
"learning_rate": 4.048e-06,
|
| 736 |
+
"loss": 2.4071,
|
| 737 |
+
"step": 2025
|
| 738 |
+
},
|
| 739 |
+
{
|
| 740 |
+
"epoch": 0.041,
|
| 741 |
+
"grad_norm": 0.7366813467336683,
|
| 742 |
+
"learning_rate": 4.098e-06,
|
| 743 |
+
"loss": 2.4113,
|
| 744 |
+
"step": 2050
|
| 745 |
+
},
|
| 746 |
+
{
|
| 747 |
+
"epoch": 0.0415,
|
| 748 |
+
"grad_norm": 0.7081408763220511,
|
| 749 |
+
"learning_rate": 4.148000000000001e-06,
|
| 750 |
+
"loss": 2.4168,
|
| 751 |
+
"step": 2075
|
| 752 |
+
},
|
| 753 |
+
{
|
| 754 |
+
"epoch": 0.042,
|
| 755 |
+
"grad_norm": 0.6912835983850483,
|
| 756 |
+
"learning_rate": 4.198e-06,
|
| 757 |
+
"loss": 2.4105,
|
| 758 |
+
"step": 2100
|
| 759 |
+
},
|
| 760 |
+
{
|
| 761 |
+
"epoch": 0.042,
|
| 762 |
+
"eval_loss": 2.438904047012329,
|
| 763 |
+
"eval_runtime": 277.7481,
|
| 764 |
+
"eval_samples_per_second": 2.96,
|
| 765 |
+
"eval_steps_per_second": 1.48,
|
| 766 |
+
"step": 2100
|
| 767 |
+
},
|
| 768 |
+
{
|
| 769 |
+
"epoch": 0.0425,
|
| 770 |
+
"grad_norm": 0.7745538733736145,
|
| 771 |
+
"learning_rate": 4.248000000000001e-06,
|
| 772 |
+
"loss": 2.4131,
|
| 773 |
+
"step": 2125
|
| 774 |
+
},
|
| 775 |
+
{
|
| 776 |
+
"epoch": 0.043,
|
| 777 |
+
"grad_norm": 0.6897576190091962,
|
| 778 |
+
"learning_rate": 4.298e-06,
|
| 779 |
+
"loss": 2.4084,
|
| 780 |
+
"step": 2150
|
| 781 |
+
},
|
| 782 |
+
{
|
| 783 |
+
"epoch": 0.0435,
|
| 784 |
+
"grad_norm": 0.7020994032566351,
|
| 785 |
+
"learning_rate": 4.3480000000000006e-06,
|
| 786 |
+
"loss": 2.4125,
|
| 787 |
+
"step": 2175
|
| 788 |
+
},
|
| 789 |
+
{
|
| 790 |
+
"epoch": 0.044,
|
| 791 |
+
"grad_norm": 0.6668651869738377,
|
| 792 |
+
"learning_rate": 4.398000000000001e-06,
|
| 793 |
+
"loss": 2.4034,
|
| 794 |
+
"step": 2200
|
| 795 |
+
},
|
| 796 |
+
{
|
| 797 |
+
"epoch": 0.044,
|
| 798 |
+
"eval_loss": 2.4380908012390137,
|
| 799 |
+
"eval_runtime": 268.2252,
|
| 800 |
+
"eval_samples_per_second": 3.065,
|
| 801 |
+
"eval_steps_per_second": 1.532,
|
| 802 |
+
"step": 2200
|
| 803 |
+
},
|
| 804 |
+
{
|
| 805 |
+
"epoch": 0.0445,
|
| 806 |
+
"grad_norm": 0.6547759047620061,
|
| 807 |
+
"learning_rate": 4.4480000000000004e-06,
|
| 808 |
+
"loss": 2.4099,
|
| 809 |
+
"step": 2225
|
| 810 |
+
},
|
| 811 |
+
{
|
| 812 |
+
"epoch": 0.045,
|
| 813 |
+
"grad_norm": 0.6865815945777785,
|
| 814 |
+
"learning_rate": 4.498e-06,
|
| 815 |
+
"loss": 2.412,
|
| 816 |
+
"step": 2250
|
| 817 |
+
},
|
| 818 |
+
{
|
| 819 |
+
"epoch": 0.0455,
|
| 820 |
+
"grad_norm": 0.6878267781655092,
|
| 821 |
+
"learning_rate": 4.548e-06,
|
| 822 |
+
"loss": 2.4137,
|
| 823 |
+
"step": 2275
|
| 824 |
+
},
|
| 825 |
+
{
|
| 826 |
+
"epoch": 0.046,
|
| 827 |
+
"grad_norm": 0.8314813616644483,
|
| 828 |
+
"learning_rate": 4.598e-06,
|
| 829 |
+
"loss": 2.4097,
|
| 830 |
+
"step": 2300
|
| 831 |
+
},
|
| 832 |
+
{
|
| 833 |
+
"epoch": 0.046,
|
| 834 |
+
"eval_loss": 2.4374496936798096,
|
| 835 |
+
"eval_runtime": 263.1701,
|
| 836 |
+
"eval_samples_per_second": 3.123,
|
| 837 |
+
"eval_steps_per_second": 1.562,
|
| 838 |
+
"step": 2300
|
| 839 |
+
},
|
| 840 |
+
{
|
| 841 |
+
"epoch": 0.0465,
|
| 842 |
+
"grad_norm": 0.6723966792931375,
|
| 843 |
+
"learning_rate": 4.648e-06,
|
| 844 |
+
"loss": 2.4051,
|
| 845 |
+
"step": 2325
|
| 846 |
+
},
|
| 847 |
+
{
|
| 848 |
+
"epoch": 0.047,
|
| 849 |
+
"grad_norm": 0.7003756914046538,
|
| 850 |
+
"learning_rate": 4.698000000000001e-06,
|
| 851 |
+
"loss": 2.4032,
|
| 852 |
+
"step": 2350
|
| 853 |
+
},
|
| 854 |
+
{
|
| 855 |
+
"epoch": 0.0475,
|
| 856 |
+
"grad_norm": 0.6747085415631567,
|
| 857 |
+
"learning_rate": 4.748e-06,
|
| 858 |
+
"loss": 2.4096,
|
| 859 |
+
"step": 2375
|
| 860 |
+
},
|
| 861 |
+
{
|
| 862 |
+
"epoch": 0.048,
|
| 863 |
+
"grad_norm": 0.6571218540079207,
|
| 864 |
+
"learning_rate": 4.7980000000000005e-06,
|
| 865 |
+
"loss": 2.4165,
|
| 866 |
+
"step": 2400
|
| 867 |
+
},
|
| 868 |
+
{
|
| 869 |
+
"epoch": 0.048,
|
| 870 |
+
"eval_loss": 2.4365923404693604,
|
| 871 |
+
"eval_runtime": 264.2268,
|
| 872 |
+
"eval_samples_per_second": 3.111,
|
| 873 |
+
"eval_steps_per_second": 1.555,
|
| 874 |
+
"step": 2400
|
| 875 |
+
},
|
| 876 |
+
{
|
| 877 |
+
"epoch": 0.0485,
|
| 878 |
+
"grad_norm": 0.7464314980483315,
|
| 879 |
+
"learning_rate": 4.848000000000001e-06,
|
| 880 |
+
"loss": 2.4098,
|
| 881 |
+
"step": 2425
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"epoch": 0.049,
|
| 885 |
+
"grad_norm": 0.6267266619200393,
|
| 886 |
+
"learning_rate": 4.898e-06,
|
| 887 |
+
"loss": 2.4019,
|
| 888 |
+
"step": 2450
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 0.0495,
|
| 892 |
+
"grad_norm": 0.6650772680412506,
|
| 893 |
+
"learning_rate": 4.948000000000001e-06,
|
| 894 |
+
"loss": 2.405,
|
| 895 |
+
"step": 2475
|
| 896 |
+
},
|
| 897 |
+
{
|
| 898 |
+
"epoch": 0.05,
|
| 899 |
+
"grad_norm": 0.7197173899674899,
|
| 900 |
+
"learning_rate": 4.998e-06,
|
| 901 |
+
"loss": 2.4095,
|
| 902 |
+
"step": 2500
|
| 903 |
+
},
|
| 904 |
+
{
|
| 905 |
+
"epoch": 0.05,
|
| 906 |
+
"eval_loss": 2.4358348846435547,
|
| 907 |
+
"eval_runtime": 266.7682,
|
| 908 |
+
"eval_samples_per_second": 3.081,
|
| 909 |
+
"eval_steps_per_second": 1.541,
|
| 910 |
+
"step": 2500
|
| 911 |
+
},
|
| 912 |
+
{
|
| 913 |
+
"epoch": 0.0505,
|
| 914 |
+
"grad_norm": 0.6249572472256157,
|
| 915 |
+
"learning_rate": 5.048000000000001e-06,
|
| 916 |
+
"loss": 2.4058,
|
| 917 |
+
"step": 2525
|
| 918 |
+
},
|
| 919 |
+
{
|
| 920 |
+
"epoch": 0.051,
|
| 921 |
+
"grad_norm": 0.7429228032719255,
|
| 922 |
+
"learning_rate": 5.098000000000001e-06,
|
| 923 |
+
"loss": 2.4084,
|
| 924 |
+
"step": 2550
|
| 925 |
+
},
|
| 926 |
+
{
|
| 927 |
+
"epoch": 0.0515,
|
| 928 |
+
"grad_norm": 0.6320325962693778,
|
| 929 |
+
"learning_rate": 5.1480000000000005e-06,
|
| 930 |
+
"loss": 2.4015,
|
| 931 |
+
"step": 2575
|
| 932 |
+
},
|
| 933 |
+
{
|
| 934 |
+
"epoch": 0.052,
|
| 935 |
+
"grad_norm": 0.672581755106835,
|
| 936 |
+
"learning_rate": 5.198000000000001e-06,
|
| 937 |
+
"loss": 2.4051,
|
| 938 |
+
"step": 2600
|
| 939 |
+
},
|
| 940 |
+
{
|
| 941 |
+
"epoch": 0.052,
|
| 942 |
+
"eval_loss": 2.4351842403411865,
|
| 943 |
+
"eval_runtime": 264.9149,
|
| 944 |
+
"eval_samples_per_second": 3.103,
|
| 945 |
+
"eval_steps_per_second": 1.551,
|
| 946 |
+
"step": 2600
|
| 947 |
+
},
|
| 948 |
+
{
|
| 949 |
+
"epoch": 0.0525,
|
| 950 |
+
"grad_norm": 0.7086480776921088,
|
| 951 |
+
"learning_rate": 5.248000000000001e-06,
|
| 952 |
+
"loss": 2.3988,
|
| 953 |
+
"step": 2625
|
| 954 |
+
},
|
| 955 |
+
{
|
| 956 |
+
"epoch": 0.053,
|
| 957 |
+
"grad_norm": 0.6774201154936552,
|
| 958 |
+
"learning_rate": 5.298000000000001e-06,
|
| 959 |
+
"loss": 2.394,
|
| 960 |
+
"step": 2650
|
| 961 |
+
},
|
| 962 |
+
{
|
| 963 |
+
"epoch": 0.0535,
|
| 964 |
+
"grad_norm": 0.6661104910300973,
|
| 965 |
+
"learning_rate": 5.348000000000001e-06,
|
| 966 |
+
"loss": 2.4034,
|
| 967 |
+
"step": 2675
|
| 968 |
+
},
|
| 969 |
+
{
|
| 970 |
+
"epoch": 0.054,
|
| 971 |
+
"grad_norm": 0.6224421593448741,
|
| 972 |
+
"learning_rate": 5.398e-06,
|
| 973 |
+
"loss": 2.3939,
|
| 974 |
+
"step": 2700
|
| 975 |
+
},
|
| 976 |
+
{
|
| 977 |
+
"epoch": 0.054,
|
| 978 |
+
"eval_loss": 2.434826374053955,
|
| 979 |
+
"eval_runtime": 264.1641,
|
| 980 |
+
"eval_samples_per_second": 3.112,
|
| 981 |
+
"eval_steps_per_second": 1.556,
|
| 982 |
+
"step": 2700
|
| 983 |
+
},
|
| 984 |
+
{
|
| 985 |
+
"epoch": 0.0545,
|
| 986 |
+
"grad_norm": 0.6944661408419767,
|
| 987 |
+
"learning_rate": 5.448e-06,
|
| 988 |
+
"loss": 2.4064,
|
| 989 |
+
"step": 2725
|
| 990 |
+
},
|
| 991 |
+
{
|
| 992 |
+
"epoch": 0.055,
|
| 993 |
+
"grad_norm": 0.6597297955298902,
|
| 994 |
+
"learning_rate": 5.498e-06,
|
| 995 |
+
"loss": 2.4051,
|
| 996 |
+
"step": 2750
|
| 997 |
+
},
|
| 998 |
+
{
|
| 999 |
+
"epoch": 0.0555,
|
| 1000 |
+
"grad_norm": 0.6526109506522182,
|
| 1001 |
+
"learning_rate": 5.548e-06,
|
| 1002 |
+
"loss": 2.4124,
|
| 1003 |
+
"step": 2775
|
| 1004 |
+
},
|
| 1005 |
+
{
|
| 1006 |
+
"epoch": 0.056,
|
| 1007 |
+
"grad_norm": 0.6528041780055424,
|
| 1008 |
+
"learning_rate": 5.5980000000000004e-06,
|
| 1009 |
+
"loss": 2.3979,
|
| 1010 |
+
"step": 2800
|
| 1011 |
+
},
|
| 1012 |
+
{
|
| 1013 |
+
"epoch": 0.056,
|
| 1014 |
+
"eval_loss": 2.4344167709350586,
|
| 1015 |
+
"eval_runtime": 264.2924,
|
| 1016 |
+
"eval_samples_per_second": 3.11,
|
| 1017 |
+
"eval_steps_per_second": 1.555,
|
| 1018 |
+
"step": 2800
|
| 1019 |
+
},
|
| 1020 |
+
{
|
| 1021 |
+
"epoch": 0.0565,
|
| 1022 |
+
"grad_norm": 0.7067565611523313,
|
| 1023 |
+
"learning_rate": 5.648e-06,
|
| 1024 |
+
"loss": 2.398,
|
| 1025 |
+
"step": 2825
|
| 1026 |
+
},
|
| 1027 |
+
{
|
| 1028 |
+
"epoch": 0.057,
|
| 1029 |
+
"grad_norm": 0.6416666495903947,
|
| 1030 |
+
"learning_rate": 5.698e-06,
|
| 1031 |
+
"loss": 2.3991,
|
| 1032 |
+
"step": 2850
|
| 1033 |
+
},
|
| 1034 |
+
{
|
| 1035 |
+
"epoch": 0.0575,
|
| 1036 |
+
"grad_norm": 0.6605105424774851,
|
| 1037 |
+
"learning_rate": 5.748e-06,
|
| 1038 |
+
"loss": 2.3962,
|
| 1039 |
+
"step": 2875
|
| 1040 |
+
},
|
| 1041 |
+
{
|
| 1042 |
+
"epoch": 0.058,
|
| 1043 |
+
"grad_norm": 0.6308761264530915,
|
| 1044 |
+
"learning_rate": 5.798e-06,
|
| 1045 |
+
"loss": 2.4058,
|
| 1046 |
+
"step": 2900
|
| 1047 |
+
},
|
| 1048 |
+
{
|
| 1049 |
+
"epoch": 0.058,
|
| 1050 |
+
"eval_loss": 2.434436082839966,
|
| 1051 |
+
"eval_runtime": 265.0112,
|
| 1052 |
+
"eval_samples_per_second": 3.102,
|
| 1053 |
+
"eval_steps_per_second": 1.551,
|
| 1054 |
+
"step": 2900
|
| 1055 |
+
},
|
| 1056 |
+
{
|
| 1057 |
+
"epoch": 0.0585,
|
| 1058 |
+
"grad_norm": 0.6363649329289001,
|
| 1059 |
+
"learning_rate": 5.848000000000001e-06,
|
| 1060 |
+
"loss": 2.3943,
|
| 1061 |
+
"step": 2925
|
| 1062 |
+
},
|
| 1063 |
+
{
|
| 1064 |
+
"epoch": 0.059,
|
| 1065 |
+
"grad_norm": 0.6147983139117156,
|
| 1066 |
+
"learning_rate": 5.898e-06,
|
| 1067 |
+
"loss": 2.3982,
|
| 1068 |
+
"step": 2950
|
| 1069 |
+
},
|
| 1070 |
+
{
|
| 1071 |
+
"epoch": 0.0595,
|
| 1072 |
+
"grad_norm": 0.611354772141602,
|
| 1073 |
+
"learning_rate": 5.9480000000000005e-06,
|
| 1074 |
+
"loss": 2.3921,
|
| 1075 |
+
"step": 2975
|
| 1076 |
+
},
|
| 1077 |
+
{
|
| 1078 |
+
"epoch": 0.06,
|
| 1079 |
+
"grad_norm": 0.6269054680170398,
|
| 1080 |
+
"learning_rate": 5.998000000000001e-06,
|
| 1081 |
+
"loss": 2.392,
|
| 1082 |
+
"step": 3000
|
| 1083 |
+
},
|
| 1084 |
+
{
|
| 1085 |
+
"epoch": 0.06,
|
| 1086 |
+
"eval_loss": 2.433990955352783,
|
| 1087 |
+
"eval_runtime": 264.2169,
|
| 1088 |
+
"eval_samples_per_second": 3.111,
|
| 1089 |
+
"eval_steps_per_second": 1.556,
|
| 1090 |
+
"step": 3000
|
| 1091 |
+
},
|
| 1092 |
+
{
|
| 1093 |
+
"epoch": 0.0605,
|
| 1094 |
+
"grad_norm": 0.6248207448228328,
|
| 1095 |
+
"learning_rate": 6.048e-06,
|
| 1096 |
+
"loss": 2.3858,
|
| 1097 |
+
"step": 3025
|
| 1098 |
+
},
|
| 1099 |
+
{
|
| 1100 |
+
"epoch": 0.061,
|
| 1101 |
+
"grad_norm": 0.6275258656299642,
|
| 1102 |
+
"learning_rate": 6.098000000000001e-06,
|
| 1103 |
+
"loss": 2.4015,
|
| 1104 |
+
"step": 3050
|
| 1105 |
+
},
|
| 1106 |
+
{
|
| 1107 |
+
"epoch": 0.0615,
|
| 1108 |
+
"grad_norm": 1.0457401571274152,
|
| 1109 |
+
"learning_rate": 6.148e-06,
|
| 1110 |
+
"loss": 2.3909,
|
| 1111 |
+
"step": 3075
|
| 1112 |
+
},
|
| 1113 |
+
{
|
| 1114 |
+
"epoch": 0.062,
|
| 1115 |
+
"grad_norm": 0.6551230863319748,
|
| 1116 |
+
"learning_rate": 6.198000000000001e-06,
|
| 1117 |
+
"loss": 2.3983,
|
| 1118 |
+
"step": 3100
|
| 1119 |
+
},
|
| 1120 |
+
{
|
| 1121 |
+
"epoch": 0.062,
|
| 1122 |
+
"eval_loss": 2.433279275894165,
|
| 1123 |
+
"eval_runtime": 264.1521,
|
| 1124 |
+
"eval_samples_per_second": 3.112,
|
| 1125 |
+
"eval_steps_per_second": 1.556,
|
| 1126 |
+
"step": 3100
|
| 1127 |
+
},
|
| 1128 |
+
{
|
| 1129 |
+
"epoch": 0.0625,
|
| 1130 |
+
"grad_norm": 0.6306746226297937,
|
| 1131 |
+
"learning_rate": 6.248000000000001e-06,
|
| 1132 |
+
"loss": 2.397,
|
| 1133 |
+
"step": 3125
|
| 1134 |
+
},
|
| 1135 |
+
{
|
| 1136 |
+
"epoch": 0.063,
|
| 1137 |
+
"grad_norm": 0.6299802316587856,
|
| 1138 |
+
"learning_rate": 6.2980000000000005e-06,
|
| 1139 |
+
"loss": 2.4018,
|
| 1140 |
+
"step": 3150
|
| 1141 |
+
},
|
| 1142 |
+
{
|
| 1143 |
+
"epoch": 0.0635,
|
| 1144 |
+
"grad_norm": 0.6265424590222634,
|
| 1145 |
+
"learning_rate": 6.348000000000001e-06,
|
| 1146 |
+
"loss": 2.4065,
|
| 1147 |
+
"step": 3175
|
| 1148 |
+
},
|
| 1149 |
+
{
|
| 1150 |
+
"epoch": 0.064,
|
| 1151 |
+
"grad_norm": 0.6717273211615455,
|
| 1152 |
+
"learning_rate": 6.398000000000001e-06,
|
| 1153 |
+
"loss": 2.3906,
|
| 1154 |
+
"step": 3200
|
| 1155 |
+
},
|
| 1156 |
+
{
|
| 1157 |
+
"epoch": 0.064,
|
| 1158 |
+
"eval_loss": 2.4333276748657227,
|
| 1159 |
+
"eval_runtime": 263.9592,
|
| 1160 |
+
"eval_samples_per_second": 3.114,
|
| 1161 |
+
"eval_steps_per_second": 1.557,
|
| 1162 |
+
"step": 3200
|
| 1163 |
+
},
|
| 1164 |
+
{
|
| 1165 |
+
"epoch": 0.0645,
|
| 1166 |
+
"grad_norm": 0.6159924635031793,
|
| 1167 |
+
"learning_rate": 6.448000000000001e-06,
|
| 1168 |
+
"loss": 2.3947,
|
| 1169 |
+
"step": 3225
|
| 1170 |
+
},
|
| 1171 |
+
{
|
| 1172 |
+
"epoch": 0.065,
|
| 1173 |
+
"grad_norm": 0.6124462043712093,
|
| 1174 |
+
"learning_rate": 6.498000000000001e-06,
|
| 1175 |
+
"loss": 2.3963,
|
| 1176 |
+
"step": 3250
|
| 1177 |
+
},
|
| 1178 |
+
{
|
| 1179 |
+
"epoch": 0.0655,
|
| 1180 |
+
"grad_norm": 0.6144378183602921,
|
| 1181 |
+
"learning_rate": 6.548000000000001e-06,
|
| 1182 |
+
"loss": 2.402,
|
| 1183 |
+
"step": 3275
|
| 1184 |
+
},
|
| 1185 |
+
{
|
| 1186 |
+
"epoch": 0.066,
|
| 1187 |
+
"grad_norm": 0.6295732934678283,
|
| 1188 |
+
"learning_rate": 6.598000000000001e-06,
|
| 1189 |
+
"loss": 2.3877,
|
| 1190 |
+
"step": 3300
|
| 1191 |
+
},
|
| 1192 |
+
{
|
| 1193 |
+
"epoch": 0.066,
|
| 1194 |
+
"eval_loss": 2.4331116676330566,
|
| 1195 |
+
"eval_runtime": 263.4524,
|
| 1196 |
+
"eval_samples_per_second": 3.12,
|
| 1197 |
+
"eval_steps_per_second": 1.56,
|
| 1198 |
+
"step": 3300
|
| 1199 |
+
},
|
| 1200 |
+
{
|
| 1201 |
+
"epoch": 0.0665,
|
| 1202 |
+
"grad_norm": 0.5938287129149346,
|
| 1203 |
+
"learning_rate": 6.648e-06,
|
| 1204 |
+
"loss": 2.389,
|
| 1205 |
+
"step": 3325
|
| 1206 |
+
},
|
| 1207 |
+
{
|
| 1208 |
+
"epoch": 0.067,
|
| 1209 |
+
"grad_norm": 0.6194783667871923,
|
| 1210 |
+
"learning_rate": 6.698e-06,
|
| 1211 |
+
"loss": 2.39,
|
| 1212 |
+
"step": 3350
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"epoch": 0.0675,
|
| 1216 |
+
"grad_norm": 0.60927231594853,
|
| 1217 |
+
"learning_rate": 6.7480000000000004e-06,
|
| 1218 |
+
"loss": 2.3968,
|
| 1219 |
+
"step": 3375
|
| 1220 |
+
},
|
| 1221 |
+
{
|
| 1222 |
+
"epoch": 0.068,
|
| 1223 |
+
"grad_norm": 0.6386175333576501,
|
| 1224 |
+
"learning_rate": 6.798e-06,
|
| 1225 |
+
"loss": 2.3861,
|
| 1226 |
+
"step": 3400
|
| 1227 |
+
},
|
| 1228 |
+
{
|
| 1229 |
+
"epoch": 0.068,
|
| 1230 |
+
"eval_loss": 2.4328911304473877,
|
| 1231 |
+
"eval_runtime": 264.2923,
|
| 1232 |
+
"eval_samples_per_second": 3.11,
|
| 1233 |
+
"eval_steps_per_second": 1.555,
|
| 1234 |
+
"step": 3400
|
| 1235 |
+
},
|
| 1236 |
+
{
|
| 1237 |
+
"epoch": 0.0685,
|
| 1238 |
+
"grad_norm": 0.6092295027577579,
|
| 1239 |
+
"learning_rate": 6.848e-06,
|
| 1240 |
+
"loss": 2.3827,
|
| 1241 |
+
"step": 3425
|
| 1242 |
+
},
|
| 1243 |
+
{
|
| 1244 |
+
"epoch": 0.069,
|
| 1245 |
+
"grad_norm": 0.5914846449422462,
|
| 1246 |
+
"learning_rate": 6.898e-06,
|
| 1247 |
+
"loss": 2.3894,
|
| 1248 |
+
"step": 3450
|
| 1249 |
+
},
|
| 1250 |
+
{
|
| 1251 |
+
"epoch": 0.0695,
|
| 1252 |
+
"grad_norm": 0.5927461214526666,
|
| 1253 |
+
"learning_rate": 6.948e-06,
|
| 1254 |
+
"loss": 2.3858,
|
| 1255 |
+
"step": 3475
|
| 1256 |
+
},
|
| 1257 |
+
{
|
| 1258 |
+
"epoch": 0.07,
|
| 1259 |
+
"grad_norm": 0.5992194088197265,
|
| 1260 |
+
"learning_rate": 6.998000000000001e-06,
|
| 1261 |
+
"loss": 2.3941,
|
| 1262 |
+
"step": 3500
|
| 1263 |
+
},
|
| 1264 |
+
{
|
| 1265 |
+
"epoch": 0.07,
|
| 1266 |
+
"eval_loss": 2.432774543762207,
|
| 1267 |
+
"eval_runtime": 263.8546,
|
| 1268 |
+
"eval_samples_per_second": 3.115,
|
| 1269 |
+
"eval_steps_per_second": 1.558,
|
| 1270 |
+
"step": 3500
|
| 1271 |
+
},
|
| 1272 |
+
{
|
| 1273 |
+
"epoch": 0.0705,
|
| 1274 |
+
"grad_norm": 0.6119297158568089,
|
| 1275 |
+
"learning_rate": 7.048e-06,
|
| 1276 |
+
"loss": 2.3897,
|
| 1277 |
+
"step": 3525
|
| 1278 |
+
},
|
| 1279 |
+
{
|
| 1280 |
+
"epoch": 0.071,
|
| 1281 |
+
"grad_norm": 0.6040666217758901,
|
| 1282 |
+
"learning_rate": 7.0980000000000005e-06,
|
| 1283 |
+
"loss": 2.3966,
|
| 1284 |
+
"step": 3550
|
| 1285 |
+
},
|
| 1286 |
+
{
|
| 1287 |
+
"epoch": 0.0715,
|
| 1288 |
+
"grad_norm": 0.6142925813030266,
|
| 1289 |
+
"learning_rate": 7.148000000000001e-06,
|
| 1290 |
+
"loss": 2.3953,
|
| 1291 |
+
"step": 3575
|
| 1292 |
+
},
|
| 1293 |
+
{
|
| 1294 |
+
"epoch": 0.072,
|
| 1295 |
+
"grad_norm": 0.5857079248330344,
|
| 1296 |
+
"learning_rate": 7.198e-06,
|
| 1297 |
+
"loss": 2.3854,
|
| 1298 |
+
"step": 3600
|
| 1299 |
+
},
|
| 1300 |
+
{
|
| 1301 |
+
"epoch": 0.072,
|
| 1302 |
+
"eval_loss": 2.432868719100952,
|
| 1303 |
+
"eval_runtime": 264.1849,
|
| 1304 |
+
"eval_samples_per_second": 3.111,
|
| 1305 |
+
"eval_steps_per_second": 1.556,
|
| 1306 |
+
"step": 3600
|
| 1307 |
+
},
|
| 1308 |
+
{
|
| 1309 |
+
"epoch": 0.0725,
|
| 1310 |
+
"grad_norm": 0.6075613052530382,
|
| 1311 |
+
"learning_rate": 7.248000000000001e-06,
|
| 1312 |
+
"loss": 2.3798,
|
| 1313 |
+
"step": 3625
|
| 1314 |
+
},
|
| 1315 |
+
{
|
| 1316 |
+
"epoch": 0.073,
|
| 1317 |
+
"grad_norm": 0.6146043204282547,
|
| 1318 |
+
"learning_rate": 7.298e-06,
|
| 1319 |
+
"loss": 2.3894,
|
| 1320 |
+
"step": 3650
|
| 1321 |
+
},
|
| 1322 |
+
{
|
| 1323 |
+
"epoch": 0.0735,
|
| 1324 |
+
"grad_norm": 0.613284002341936,
|
| 1325 |
+
"learning_rate": 7.348000000000001e-06,
|
| 1326 |
+
"loss": 2.3897,
|
| 1327 |
+
"step": 3675
|
| 1328 |
+
},
|
| 1329 |
+
{
|
| 1330 |
+
"epoch": 0.074,
|
| 1331 |
+
"grad_norm": 0.6694404263159593,
|
| 1332 |
+
"learning_rate": 7.398000000000001e-06,
|
| 1333 |
+
"loss": 2.3925,
|
| 1334 |
+
"step": 3700
|
| 1335 |
+
},
|
| 1336 |
+
{
|
| 1337 |
+
"epoch": 0.074,
|
| 1338 |
+
"eval_loss": 2.4324021339416504,
|
| 1339 |
+
"eval_runtime": 263.3107,
|
| 1340 |
+
"eval_samples_per_second": 3.122,
|
| 1341 |
+
"eval_steps_per_second": 1.561,
|
| 1342 |
+
"step": 3700
|
| 1343 |
+
},
|
| 1344 |
+
{
|
| 1345 |
+
"epoch": 0.0745,
|
| 1346 |
+
"grad_norm": 0.5756401973694445,
|
| 1347 |
+
"learning_rate": 7.4480000000000005e-06,
|
| 1348 |
+
"loss": 2.3894,
|
| 1349 |
+
"step": 3725
|
| 1350 |
+
},
|
| 1351 |
+
{
|
| 1352 |
+
"epoch": 0.075,
|
| 1353 |
+
"grad_norm": 0.5945783703417461,
|
| 1354 |
+
"learning_rate": 7.498000000000001e-06,
|
| 1355 |
+
"loss": 2.3928,
|
| 1356 |
+
"step": 3750
|
| 1357 |
+
},
|
| 1358 |
+
{
|
| 1359 |
+
"epoch": 0.0755,
|
| 1360 |
+
"grad_norm": 0.5935750222986942,
|
| 1361 |
+
"learning_rate": 7.548000000000001e-06,
|
| 1362 |
+
"loss": 2.3774,
|
| 1363 |
+
"step": 3775
|
| 1364 |
+
},
|
| 1365 |
+
{
|
| 1366 |
+
"epoch": 0.076,
|
| 1367 |
+
"grad_norm": 0.5938734543073783,
|
| 1368 |
+
"learning_rate": 7.598000000000001e-06,
|
| 1369 |
+
"loss": 2.3776,
|
| 1370 |
+
"step": 3800
|
| 1371 |
+
},
|
| 1372 |
+
{
|
| 1373 |
+
"epoch": 0.076,
|
| 1374 |
+
"eval_loss": 2.432751178741455,
|
| 1375 |
+
"eval_runtime": 263.8929,
|
| 1376 |
+
"eval_samples_per_second": 3.115,
|
| 1377 |
+
"eval_steps_per_second": 1.557,
|
| 1378 |
+
"step": 3800
|
| 1379 |
+
},
|
| 1380 |
+
{
|
| 1381 |
+
"epoch": 0.0765,
|
| 1382 |
+
"grad_norm": 0.595820899700728,
|
| 1383 |
+
"learning_rate": 7.648e-06,
|
| 1384 |
+
"loss": 2.3804,
|
| 1385 |
+
"step": 3825
|
| 1386 |
+
},
|
| 1387 |
+
{
|
| 1388 |
+
"epoch": 0.077,
|
| 1389 |
+
"grad_norm": 0.6079304106413467,
|
| 1390 |
+
"learning_rate": 7.698000000000002e-06,
|
| 1391 |
+
"loss": 2.3917,
|
| 1392 |
+
"step": 3850
|
| 1393 |
+
},
|
| 1394 |
+
{
|
| 1395 |
+
"epoch": 0.0775,
|
| 1396 |
+
"grad_norm": 0.6083448146618482,
|
| 1397 |
+
"learning_rate": 7.748000000000001e-06,
|
| 1398 |
+
"loss": 2.3842,
|
| 1399 |
+
"step": 3875
|
| 1400 |
+
},
|
| 1401 |
+
{
|
| 1402 |
+
"epoch": 0.078,
|
| 1403 |
+
"grad_norm": 0.6128893415605828,
|
| 1404 |
+
"learning_rate": 7.798e-06,
|
| 1405 |
+
"loss": 2.3806,
|
| 1406 |
+
"step": 3900
|
| 1407 |
+
},
|
| 1408 |
+
{
|
| 1409 |
+
"epoch": 0.078,
|
| 1410 |
+
"eval_loss": 2.4325239658355713,
|
| 1411 |
+
"eval_runtime": 263.6693,
|
| 1412 |
+
"eval_samples_per_second": 3.118,
|
| 1413 |
+
"eval_steps_per_second": 1.559,
|
| 1414 |
+
"step": 3900
|
| 1415 |
+
},
|
| 1416 |
+
{
|
| 1417 |
+
"epoch": 0.0785,
|
| 1418 |
+
"grad_norm": 0.6079041195191952,
|
| 1419 |
+
"learning_rate": 7.848000000000002e-06,
|
| 1420 |
+
"loss": 2.3801,
|
| 1421 |
+
"step": 3925
|
| 1422 |
+
},
|
| 1423 |
+
{
|
| 1424 |
+
"epoch": 0.079,
|
| 1425 |
+
"grad_norm": 0.6075689821557235,
|
| 1426 |
+
"learning_rate": 7.898e-06,
|
| 1427 |
+
"loss": 2.3797,
|
| 1428 |
+
"step": 3950
|
| 1429 |
+
},
|
| 1430 |
+
{
|
| 1431 |
+
"epoch": 0.0795,
|
| 1432 |
+
"grad_norm": 0.5882326737716994,
|
| 1433 |
+
"learning_rate": 7.948e-06,
|
| 1434 |
+
"loss": 2.3905,
|
| 1435 |
+
"step": 3975
|
| 1436 |
+
},
|
| 1437 |
+
{
|
| 1438 |
+
"epoch": 0.08,
|
| 1439 |
+
"grad_norm": 0.5828476462223788,
|
| 1440 |
+
"learning_rate": 7.998e-06,
|
| 1441 |
+
"loss": 2.3806,
|
| 1442 |
+
"step": 4000
|
| 1443 |
+
},
|
| 1444 |
+
{
|
| 1445 |
+
"epoch": 0.08,
|
| 1446 |
+
"eval_loss": 2.4323527812957764,
|
| 1447 |
+
"eval_runtime": 263.9786,
|
| 1448 |
+
"eval_samples_per_second": 3.114,
|
| 1449 |
+
"eval_steps_per_second": 1.557,
|
| 1450 |
+
"step": 4000
|
| 1451 |
}
|
| 1452 |
],
|
| 1453 |
"logging_steps": 25,
|
|
|
|
| 1467 |
"attributes": {}
|
| 1468 |
}
|
| 1469 |
},
|
| 1470 |
+
"total_flos": 1.2732809193308815e+19,
|
| 1471 |
"train_batch_size": 1,
|
| 1472 |
"trial_name": null,
|
| 1473 |
"trial_params": null
|