ocr version 1
Browse files- trainer_state.json +387 -0
trainer_state.json
ADDED
|
@@ -0,0 +1,387 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": null,
|
| 3 |
+
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 5.0,
|
| 5 |
+
"eval_steps": 500,
|
| 6 |
+
"global_step": 1560,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.0,
|
| 13 |
+
"grad_norm": 134.19052124023438,
|
| 14 |
+
"learning_rate": 4.996794871794872e-05,
|
| 15 |
+
"loss": 9.1177,
|
| 16 |
+
"step": 1
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 0.1,
|
| 20 |
+
"grad_norm": 46.553192138671875,
|
| 21 |
+
"learning_rate": 4.9006410256410256e-05,
|
| 22 |
+
"loss": 2.6116,
|
| 23 |
+
"step": 31
|
| 24 |
+
},
|
| 25 |
+
{
|
| 26 |
+
"epoch": 0.2,
|
| 27 |
+
"grad_norm": 17.68166732788086,
|
| 28 |
+
"learning_rate": 4.8012820512820516e-05,
|
| 29 |
+
"loss": 0.8879,
|
| 30 |
+
"step": 62
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"epoch": 0.3,
|
| 34 |
+
"grad_norm": 17.172128677368164,
|
| 35 |
+
"learning_rate": 4.701923076923077e-05,
|
| 36 |
+
"loss": 1.0567,
|
| 37 |
+
"step": 93
|
| 38 |
+
},
|
| 39 |
+
{
|
| 40 |
+
"epoch": 0.4,
|
| 41 |
+
"grad_norm": 104.65974426269531,
|
| 42 |
+
"learning_rate": 4.602564102564102e-05,
|
| 43 |
+
"loss": 0.9872,
|
| 44 |
+
"step": 124
|
| 45 |
+
},
|
| 46 |
+
{
|
| 47 |
+
"epoch": 0.5,
|
| 48 |
+
"grad_norm": 16.036008834838867,
|
| 49 |
+
"learning_rate": 4.503205128205128e-05,
|
| 50 |
+
"loss": 0.9717,
|
| 51 |
+
"step": 155
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 0.6,
|
| 55 |
+
"grad_norm": 27.394895553588867,
|
| 56 |
+
"learning_rate": 4.403846153846154e-05,
|
| 57 |
+
"loss": 0.8679,
|
| 58 |
+
"step": 186
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"epoch": 0.7,
|
| 62 |
+
"grad_norm": 37.157169342041016,
|
| 63 |
+
"learning_rate": 4.30448717948718e-05,
|
| 64 |
+
"loss": 0.8257,
|
| 65 |
+
"step": 217
|
| 66 |
+
},
|
| 67 |
+
{
|
| 68 |
+
"epoch": 0.79,
|
| 69 |
+
"grad_norm": 8.90817928314209,
|
| 70 |
+
"learning_rate": 4.205128205128206e-05,
|
| 71 |
+
"loss": 0.8604,
|
| 72 |
+
"step": 248
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"epoch": 0.89,
|
| 76 |
+
"grad_norm": 13.38546085357666,
|
| 77 |
+
"learning_rate": 4.105769230769231e-05,
|
| 78 |
+
"loss": 0.8587,
|
| 79 |
+
"step": 279
|
| 80 |
+
},
|
| 81 |
+
{
|
| 82 |
+
"epoch": 0.99,
|
| 83 |
+
"grad_norm": 32.03287124633789,
|
| 84 |
+
"learning_rate": 4.006410256410257e-05,
|
| 85 |
+
"loss": 0.8512,
|
| 86 |
+
"step": 310
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"epoch": 1.09,
|
| 90 |
+
"grad_norm": 20.29245376586914,
|
| 91 |
+
"learning_rate": 3.9070512820512824e-05,
|
| 92 |
+
"loss": 0.6728,
|
| 93 |
+
"step": 341
|
| 94 |
+
},
|
| 95 |
+
{
|
| 96 |
+
"epoch": 1.19,
|
| 97 |
+
"grad_norm": 10.910124778747559,
|
| 98 |
+
"learning_rate": 3.807692307692308e-05,
|
| 99 |
+
"loss": 0.611,
|
| 100 |
+
"step": 372
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 1.29,
|
| 104 |
+
"grad_norm": 14.703102111816406,
|
| 105 |
+
"learning_rate": 3.708333333333334e-05,
|
| 106 |
+
"loss": 0.6468,
|
| 107 |
+
"step": 403
|
| 108 |
+
},
|
| 109 |
+
{
|
| 110 |
+
"epoch": 1.39,
|
| 111 |
+
"grad_norm": 9.841747283935547,
|
| 112 |
+
"learning_rate": 3.608974358974359e-05,
|
| 113 |
+
"loss": 0.5327,
|
| 114 |
+
"step": 434
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 1.49,
|
| 118 |
+
"grad_norm": 15.09598445892334,
|
| 119 |
+
"learning_rate": 3.5096153846153845e-05,
|
| 120 |
+
"loss": 0.5669,
|
| 121 |
+
"step": 465
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"epoch": 1.59,
|
| 125 |
+
"grad_norm": 18.652040481567383,
|
| 126 |
+
"learning_rate": 3.4102564102564105e-05,
|
| 127 |
+
"loss": 0.4854,
|
| 128 |
+
"step": 496
|
| 129 |
+
},
|
| 130 |
+
{
|
| 131 |
+
"epoch": 1.69,
|
| 132 |
+
"grad_norm": 8.733405113220215,
|
| 133 |
+
"learning_rate": 3.310897435897436e-05,
|
| 134 |
+
"loss": 0.5627,
|
| 135 |
+
"step": 527
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"epoch": 1.79,
|
| 139 |
+
"grad_norm": 21.200965881347656,
|
| 140 |
+
"learning_rate": 3.211538461538462e-05,
|
| 141 |
+
"loss": 0.5001,
|
| 142 |
+
"step": 558
|
| 143 |
+
},
|
| 144 |
+
{
|
| 145 |
+
"epoch": 1.89,
|
| 146 |
+
"grad_norm": 7.368778228759766,
|
| 147 |
+
"learning_rate": 3.112179487179487e-05,
|
| 148 |
+
"loss": 0.4897,
|
| 149 |
+
"step": 589
|
| 150 |
+
},
|
| 151 |
+
{
|
| 152 |
+
"epoch": 1.99,
|
| 153 |
+
"grad_norm": 16.03817367553711,
|
| 154 |
+
"learning_rate": 3.012820512820513e-05,
|
| 155 |
+
"loss": 0.5294,
|
| 156 |
+
"step": 620
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"epoch": 2.09,
|
| 160 |
+
"grad_norm": 35.705631256103516,
|
| 161 |
+
"learning_rate": 2.913461538461539e-05,
|
| 162 |
+
"loss": 0.4197,
|
| 163 |
+
"step": 651
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"epoch": 2.19,
|
| 167 |
+
"grad_norm": 10.462698936462402,
|
| 168 |
+
"learning_rate": 2.8141025641025643e-05,
|
| 169 |
+
"loss": 0.3589,
|
| 170 |
+
"step": 682
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"epoch": 2.29,
|
| 174 |
+
"grad_norm": 7.011005878448486,
|
| 175 |
+
"learning_rate": 2.7147435897435896e-05,
|
| 176 |
+
"loss": 0.4396,
|
| 177 |
+
"step": 713
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 2.38,
|
| 181 |
+
"grad_norm": 4.32899284362793,
|
| 182 |
+
"learning_rate": 2.6153846153846157e-05,
|
| 183 |
+
"loss": 0.4604,
|
| 184 |
+
"step": 744
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"epoch": 2.48,
|
| 188 |
+
"grad_norm": 5.010735988616943,
|
| 189 |
+
"learning_rate": 2.516025641025641e-05,
|
| 190 |
+
"loss": 0.3764,
|
| 191 |
+
"step": 775
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 2.58,
|
| 195 |
+
"grad_norm": 6.4428582191467285,
|
| 196 |
+
"learning_rate": 2.4166666666666667e-05,
|
| 197 |
+
"loss": 0.3035,
|
| 198 |
+
"step": 806
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"epoch": 2.68,
|
| 202 |
+
"grad_norm": 11.917937278747559,
|
| 203 |
+
"learning_rate": 2.3173076923076924e-05,
|
| 204 |
+
"loss": 0.3287,
|
| 205 |
+
"step": 837
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 2.78,
|
| 209 |
+
"grad_norm": 18.710346221923828,
|
| 210 |
+
"learning_rate": 2.217948717948718e-05,
|
| 211 |
+
"loss": 0.3209,
|
| 212 |
+
"step": 868
|
| 213 |
+
},
|
| 214 |
+
{
|
| 215 |
+
"epoch": 2.88,
|
| 216 |
+
"grad_norm": 5.8296990394592285,
|
| 217 |
+
"learning_rate": 2.1185897435897437e-05,
|
| 218 |
+
"loss": 0.329,
|
| 219 |
+
"step": 899
|
| 220 |
+
},
|
| 221 |
+
{
|
| 222 |
+
"epoch": 2.98,
|
| 223 |
+
"grad_norm": 11.532828330993652,
|
| 224 |
+
"learning_rate": 2.0192307692307694e-05,
|
| 225 |
+
"loss": 0.2934,
|
| 226 |
+
"step": 930
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"epoch": 3.08,
|
| 230 |
+
"grad_norm": 15.149397850036621,
|
| 231 |
+
"learning_rate": 1.919871794871795e-05,
|
| 232 |
+
"loss": 0.2682,
|
| 233 |
+
"step": 961
|
| 234 |
+
},
|
| 235 |
+
{
|
| 236 |
+
"epoch": 3.18,
|
| 237 |
+
"grad_norm": 3.612698793411255,
|
| 238 |
+
"learning_rate": 1.8205128205128204e-05,
|
| 239 |
+
"loss": 0.264,
|
| 240 |
+
"step": 992
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"epoch": 3.28,
|
| 244 |
+
"grad_norm": 1.7986979484558105,
|
| 245 |
+
"learning_rate": 1.721153846153846e-05,
|
| 246 |
+
"loss": 0.2085,
|
| 247 |
+
"step": 1023
|
| 248 |
+
},
|
| 249 |
+
{
|
| 250 |
+
"epoch": 3.38,
|
| 251 |
+
"grad_norm": 3.6060192584991455,
|
| 252 |
+
"learning_rate": 1.6217948717948718e-05,
|
| 253 |
+
"loss": 0.247,
|
| 254 |
+
"step": 1054
|
| 255 |
+
},
|
| 256 |
+
{
|
| 257 |
+
"epoch": 3.48,
|
| 258 |
+
"grad_norm": 2.7116451263427734,
|
| 259 |
+
"learning_rate": 1.5224358974358973e-05,
|
| 260 |
+
"loss": 0.2104,
|
| 261 |
+
"step": 1085
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"epoch": 3.58,
|
| 265 |
+
"grad_norm": 4.838766574859619,
|
| 266 |
+
"learning_rate": 1.423076923076923e-05,
|
| 267 |
+
"loss": 0.234,
|
| 268 |
+
"step": 1116
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 3.68,
|
| 272 |
+
"grad_norm": 2.237657070159912,
|
| 273 |
+
"learning_rate": 1.3237179487179489e-05,
|
| 274 |
+
"loss": 0.2033,
|
| 275 |
+
"step": 1147
|
| 276 |
+
},
|
| 277 |
+
{
|
| 278 |
+
"epoch": 3.78,
|
| 279 |
+
"grad_norm": 1.6461944580078125,
|
| 280 |
+
"learning_rate": 1.2243589743589744e-05,
|
| 281 |
+
"loss": 0.2098,
|
| 282 |
+
"step": 1178
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 3.88,
|
| 286 |
+
"grad_norm": 6.327276229858398,
|
| 287 |
+
"learning_rate": 1.125e-05,
|
| 288 |
+
"loss": 0.2073,
|
| 289 |
+
"step": 1209
|
| 290 |
+
},
|
| 291 |
+
{
|
| 292 |
+
"epoch": 3.97,
|
| 293 |
+
"grad_norm": 2.9778146743774414,
|
| 294 |
+
"learning_rate": 1.0256410256410256e-05,
|
| 295 |
+
"loss": 0.1988,
|
| 296 |
+
"step": 1240
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"epoch": 4.07,
|
| 300 |
+
"grad_norm": 1.4347281455993652,
|
| 301 |
+
"learning_rate": 9.262820512820514e-06,
|
| 302 |
+
"loss": 0.1664,
|
| 303 |
+
"step": 1271
|
| 304 |
+
},
|
| 305 |
+
{
|
| 306 |
+
"epoch": 4.17,
|
| 307 |
+
"grad_norm": 2.844505786895752,
|
| 308 |
+
"learning_rate": 8.26923076923077e-06,
|
| 309 |
+
"loss": 0.1529,
|
| 310 |
+
"step": 1302
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 4.27,
|
| 314 |
+
"grad_norm": 1.985013723373413,
|
| 315 |
+
"learning_rate": 7.275641025641026e-06,
|
| 316 |
+
"loss": 0.1447,
|
| 317 |
+
"step": 1333
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 4.37,
|
| 321 |
+
"grad_norm": 2.9127843379974365,
|
| 322 |
+
"learning_rate": 6.282051282051282e-06,
|
| 323 |
+
"loss": 0.1375,
|
| 324 |
+
"step": 1364
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 4.47,
|
| 328 |
+
"grad_norm": 2.6174566745758057,
|
| 329 |
+
"learning_rate": 5.288461538461538e-06,
|
| 330 |
+
"loss": 0.1515,
|
| 331 |
+
"step": 1395
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 4.57,
|
| 335 |
+
"grad_norm": 1.2411088943481445,
|
| 336 |
+
"learning_rate": 4.294871794871795e-06,
|
| 337 |
+
"loss": 0.1408,
|
| 338 |
+
"step": 1426
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 4.67,
|
| 342 |
+
"grad_norm": 1.8333454132080078,
|
| 343 |
+
"learning_rate": 3.3012820512820517e-06,
|
| 344 |
+
"loss": 0.1372,
|
| 345 |
+
"step": 1457
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 4.77,
|
| 349 |
+
"grad_norm": 1.785672903060913,
|
| 350 |
+
"learning_rate": 2.307692307692308e-06,
|
| 351 |
+
"loss": 0.1409,
|
| 352 |
+
"step": 1488
|
| 353 |
+
},
|
| 354 |
+
{
|
| 355 |
+
"epoch": 4.87,
|
| 356 |
+
"grad_norm": 3.533236026763916,
|
| 357 |
+
"learning_rate": 1.3141025641025643e-06,
|
| 358 |
+
"loss": 0.1276,
|
| 359 |
+
"step": 1519
|
| 360 |
+
},
|
| 361 |
+
{
|
| 362 |
+
"epoch": 4.97,
|
| 363 |
+
"grad_norm": 1.3145009279251099,
|
| 364 |
+
"learning_rate": 3.205128205128205e-07,
|
| 365 |
+
"loss": 0.1329,
|
| 366 |
+
"step": 1550
|
| 367 |
+
},
|
| 368 |
+
{
|
| 369 |
+
"epoch": 5.0,
|
| 370 |
+
"step": 1560,
|
| 371 |
+
"total_flos": 1.845867535870722e+19,
|
| 372 |
+
"train_loss": 0.4759287901413746,
|
| 373 |
+
"train_runtime": 3868.1419,
|
| 374 |
+
"train_samples_per_second": 3.224,
|
| 375 |
+
"train_steps_per_second": 0.403
|
| 376 |
+
}
|
| 377 |
+
],
|
| 378 |
+
"logging_steps": 31,
|
| 379 |
+
"max_steps": 1560,
|
| 380 |
+
"num_input_tokens_seen": 0,
|
| 381 |
+
"num_train_epochs": 5,
|
| 382 |
+
"save_steps": 500,
|
| 383 |
+
"total_flos": 1.845867535870722e+19,
|
| 384 |
+
"train_batch_size": 8,
|
| 385 |
+
"trial_name": null,
|
| 386 |
+
"trial_params": null
|
| 387 |
+
}
|