update to 5K steps
Browse files- model-00001-of-00002.safetensors +1 -1
- model-00002-of-00002.safetensors +1 -1
- optimizer.pt +1 -1
- rng_state.pth +2 -2
- scheduler.pt +1 -1
- trainer_state.json +347 -3
model-00001-of-00002.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 4986380064
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5e978119ee425405b19afc309b1ac637b507f9a95b931ad8989930e197cdea18
|
| 3 |
size 4986380064
|
model-00002-of-00002.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 399532808
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:eb38cf9ee1edb794d18560d40bb29e829f95856fde9b87466cd6b8526e2ee450
|
| 3 |
size 399532808
|
optimizer.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 2699039674
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6582836e4bc0c137ecd1dd128ed86ee2a49ffa2f6655159a311931e73de33b01
|
| 3 |
size 2699039674
|
rng_state.pth
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 sha256:1fddbe326fadb3ce8c449c2642e8a1dd894d5eabee5fd4473ac2c38deccaf19b
|
| 3 |
+
size 14244
|
scheduler.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 1064
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2be67f6aac8e482bb2022409709d8774ffb125292c0c9cf025c0ae747f3a6d57
|
| 3 |
size 1064
|
trainer_state.json
CHANGED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
{
|
| 2 |
"best_metric": null,
|
| 3 |
"best_model_checkpoint": null,
|
| 4 |
-
"epoch":
|
| 5 |
"eval_steps": 500,
|
| 6 |
-
"global_step":
|
| 7 |
"is_hyper_param_search": false,
|
| 8 |
"is_local_process_zero": true,
|
| 9 |
"is_world_process_zero": true,
|
|
@@ -93,6 +93,350 @@
|
|
| 93 |
"eval_samples_per_second": 15.394,
|
| 94 |
"eval_steps_per_second": 3.848,
|
| 95 |
"step": 1000
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 96 |
}
|
| 97 |
],
|
| 98 |
"logging_steps": 100,
|
|
@@ -100,7 +444,7 @@
|
|
| 100 |
"num_input_tokens_seen": 0,
|
| 101 |
"num_train_epochs": 9,
|
| 102 |
"save_steps": 500,
|
| 103 |
-
"total_flos":
|
| 104 |
"train_batch_size": 4,
|
| 105 |
"trial_name": null,
|
| 106 |
"trial_params": null
|
|
|
|
| 1 |
{
|
| 2 |
"best_metric": null,
|
| 3 |
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 8.605851979345955,
|
| 5 |
"eval_steps": 500,
|
| 6 |
+
"global_step": 5000,
|
| 7 |
"is_hyper_param_search": false,
|
| 8 |
"is_local_process_zero": true,
|
| 9 |
"is_world_process_zero": true,
|
|
|
|
| 93 |
"eval_samples_per_second": 15.394,
|
| 94 |
"eval_steps_per_second": 3.848,
|
| 95 |
"step": 1000
|
| 96 |
+
},
|
| 97 |
+
{
|
| 98 |
+
"epoch": 1.89,
|
| 99 |
+
"grad_norm": 0.941031813621521,
|
| 100 |
+
"learning_rate": 1.7333333333333336e-05,
|
| 101 |
+
"loss": 0.2556,
|
| 102 |
+
"step": 1100
|
| 103 |
+
},
|
| 104 |
+
{
|
| 105 |
+
"epoch": 2.07,
|
| 106 |
+
"grad_norm": 1.5538479089736938,
|
| 107 |
+
"learning_rate": 1.688888888888889e-05,
|
| 108 |
+
"loss": 0.2188,
|
| 109 |
+
"step": 1200
|
| 110 |
+
},
|
| 111 |
+
{
|
| 112 |
+
"epoch": 2.24,
|
| 113 |
+
"grad_norm": 1.7369763851165771,
|
| 114 |
+
"learning_rate": 1.6444444444444444e-05,
|
| 115 |
+
"loss": 0.1589,
|
| 116 |
+
"step": 1300
|
| 117 |
+
},
|
| 118 |
+
{
|
| 119 |
+
"epoch": 2.41,
|
| 120 |
+
"grad_norm": 1.8593116998672485,
|
| 121 |
+
"learning_rate": 1.6000000000000003e-05,
|
| 122 |
+
"loss": 0.1526,
|
| 123 |
+
"step": 1400
|
| 124 |
+
},
|
| 125 |
+
{
|
| 126 |
+
"epoch": 2.58,
|
| 127 |
+
"grad_norm": 0.9025155305862427,
|
| 128 |
+
"learning_rate": 1.555555555555556e-05,
|
| 129 |
+
"loss": 0.1512,
|
| 130 |
+
"step": 1500
|
| 131 |
+
},
|
| 132 |
+
{
|
| 133 |
+
"epoch": 2.58,
|
| 134 |
+
"eval_loss": 0.23454324901103973,
|
| 135 |
+
"eval_runtime": 64.9882,
|
| 136 |
+
"eval_samples_per_second": 15.387,
|
| 137 |
+
"eval_steps_per_second": 3.847,
|
| 138 |
+
"step": 1500
|
| 139 |
+
},
|
| 140 |
+
{
|
| 141 |
+
"epoch": 2.75,
|
| 142 |
+
"grad_norm": 1.2440764904022217,
|
| 143 |
+
"learning_rate": 1.5111111111111112e-05,
|
| 144 |
+
"loss": 0.1596,
|
| 145 |
+
"step": 1600
|
| 146 |
+
},
|
| 147 |
+
{
|
| 148 |
+
"epoch": 2.93,
|
| 149 |
+
"grad_norm": 0.9365782141685486,
|
| 150 |
+
"learning_rate": 1.4666666666666666e-05,
|
| 151 |
+
"loss": 0.1603,
|
| 152 |
+
"step": 1700
|
| 153 |
+
},
|
| 154 |
+
{
|
| 155 |
+
"epoch": 3.1,
|
| 156 |
+
"grad_norm": 0.977448582649231,
|
| 157 |
+
"learning_rate": 1.4222222222222224e-05,
|
| 158 |
+
"loss": 0.1086,
|
| 159 |
+
"step": 1800
|
| 160 |
+
},
|
| 161 |
+
{
|
| 162 |
+
"epoch": 3.27,
|
| 163 |
+
"grad_norm": 0.9493745565414429,
|
| 164 |
+
"learning_rate": 1.377777777777778e-05,
|
| 165 |
+
"loss": 0.0822,
|
| 166 |
+
"step": 1900
|
| 167 |
+
},
|
| 168 |
+
{
|
| 169 |
+
"epoch": 3.44,
|
| 170 |
+
"grad_norm": 1.437554955482483,
|
| 171 |
+
"learning_rate": 1.3333333333333333e-05,
|
| 172 |
+
"loss": 0.0838,
|
| 173 |
+
"step": 2000
|
| 174 |
+
},
|
| 175 |
+
{
|
| 176 |
+
"epoch": 3.44,
|
| 177 |
+
"eval_loss": 0.258994460105896,
|
| 178 |
+
"eval_runtime": 65.0096,
|
| 179 |
+
"eval_samples_per_second": 15.382,
|
| 180 |
+
"eval_steps_per_second": 3.846,
|
| 181 |
+
"step": 2000
|
| 182 |
+
},
|
| 183 |
+
{
|
| 184 |
+
"epoch": 3.61,
|
| 185 |
+
"grad_norm": 1.1978514194488525,
|
| 186 |
+
"learning_rate": 1.288888888888889e-05,
|
| 187 |
+
"loss": 0.0849,
|
| 188 |
+
"step": 2100
|
| 189 |
+
},
|
| 190 |
+
{
|
| 191 |
+
"epoch": 3.79,
|
| 192 |
+
"grad_norm": 1.1100072860717773,
|
| 193 |
+
"learning_rate": 1.2444444444444446e-05,
|
| 194 |
+
"loss": 0.0822,
|
| 195 |
+
"step": 2200
|
| 196 |
+
},
|
| 197 |
+
{
|
| 198 |
+
"epoch": 3.96,
|
| 199 |
+
"grad_norm": 1.0157575607299805,
|
| 200 |
+
"learning_rate": 1.2e-05,
|
| 201 |
+
"loss": 0.0799,
|
| 202 |
+
"step": 2300
|
| 203 |
+
},
|
| 204 |
+
{
|
| 205 |
+
"epoch": 4.13,
|
| 206 |
+
"grad_norm": 1.1923385858535767,
|
| 207 |
+
"learning_rate": 1.1555555555555556e-05,
|
| 208 |
+
"loss": 0.05,
|
| 209 |
+
"step": 2400
|
| 210 |
+
},
|
| 211 |
+
{
|
| 212 |
+
"epoch": 4.3,
|
| 213 |
+
"grad_norm": 1.1894739866256714,
|
| 214 |
+
"learning_rate": 1.1111111111111113e-05,
|
| 215 |
+
"loss": 0.0401,
|
| 216 |
+
"step": 2500
|
| 217 |
+
},
|
| 218 |
+
{
|
| 219 |
+
"epoch": 4.3,
|
| 220 |
+
"eval_loss": 0.3230312466621399,
|
| 221 |
+
"eval_runtime": 65.0356,
|
| 222 |
+
"eval_samples_per_second": 15.376,
|
| 223 |
+
"eval_steps_per_second": 3.844,
|
| 224 |
+
"step": 2500
|
| 225 |
+
},
|
| 226 |
+
{
|
| 227 |
+
"epoch": 4.48,
|
| 228 |
+
"grad_norm": 1.5902525186538696,
|
| 229 |
+
"learning_rate": 1.0666666666666667e-05,
|
| 230 |
+
"loss": 0.0404,
|
| 231 |
+
"step": 2600
|
| 232 |
+
},
|
| 233 |
+
{
|
| 234 |
+
"epoch": 4.65,
|
| 235 |
+
"grad_norm": 1.3404467105865479,
|
| 236 |
+
"learning_rate": 1.0222222222222223e-05,
|
| 237 |
+
"loss": 0.0418,
|
| 238 |
+
"step": 2700
|
| 239 |
+
},
|
| 240 |
+
{
|
| 241 |
+
"epoch": 4.82,
|
| 242 |
+
"grad_norm": 1.4325445890426636,
|
| 243 |
+
"learning_rate": 9.777777777777779e-06,
|
| 244 |
+
"loss": 0.0441,
|
| 245 |
+
"step": 2800
|
| 246 |
+
},
|
| 247 |
+
{
|
| 248 |
+
"epoch": 4.99,
|
| 249 |
+
"grad_norm": 1.3849271535873413,
|
| 250 |
+
"learning_rate": 9.333333333333334e-06,
|
| 251 |
+
"loss": 0.0429,
|
| 252 |
+
"step": 2900
|
| 253 |
+
},
|
| 254 |
+
{
|
| 255 |
+
"epoch": 5.16,
|
| 256 |
+
"grad_norm": 1.0262218713760376,
|
| 257 |
+
"learning_rate": 8.888888888888888e-06,
|
| 258 |
+
"loss": 0.0196,
|
| 259 |
+
"step": 3000
|
| 260 |
+
},
|
| 261 |
+
{
|
| 262 |
+
"epoch": 5.16,
|
| 263 |
+
"eval_loss": 0.38104376196861267,
|
| 264 |
+
"eval_runtime": 64.9932,
|
| 265 |
+
"eval_samples_per_second": 15.386,
|
| 266 |
+
"eval_steps_per_second": 3.847,
|
| 267 |
+
"step": 3000
|
| 268 |
+
},
|
| 269 |
+
{
|
| 270 |
+
"epoch": 5.34,
|
| 271 |
+
"grad_norm": 1.4700788259506226,
|
| 272 |
+
"learning_rate": 8.444444444444446e-06,
|
| 273 |
+
"loss": 0.0205,
|
| 274 |
+
"step": 3100
|
| 275 |
+
},
|
| 276 |
+
{
|
| 277 |
+
"epoch": 5.51,
|
| 278 |
+
"grad_norm": 0.9892378449440002,
|
| 279 |
+
"learning_rate": 8.000000000000001e-06,
|
| 280 |
+
"loss": 0.0207,
|
| 281 |
+
"step": 3200
|
| 282 |
+
},
|
| 283 |
+
{
|
| 284 |
+
"epoch": 5.68,
|
| 285 |
+
"grad_norm": 0.4820214509963989,
|
| 286 |
+
"learning_rate": 7.555555555555556e-06,
|
| 287 |
+
"loss": 0.0204,
|
| 288 |
+
"step": 3300
|
| 289 |
+
},
|
| 290 |
+
{
|
| 291 |
+
"epoch": 5.85,
|
| 292 |
+
"grad_norm": 1.2511119842529297,
|
| 293 |
+
"learning_rate": 7.111111111111112e-06,
|
| 294 |
+
"loss": 0.0228,
|
| 295 |
+
"step": 3400
|
| 296 |
+
},
|
| 297 |
+
{
|
| 298 |
+
"epoch": 6.02,
|
| 299 |
+
"grad_norm": 0.32089540362358093,
|
| 300 |
+
"learning_rate": 6.666666666666667e-06,
|
| 301 |
+
"loss": 0.0197,
|
| 302 |
+
"step": 3500
|
| 303 |
+
},
|
| 304 |
+
{
|
| 305 |
+
"epoch": 6.02,
|
| 306 |
+
"eval_loss": 0.3612504303455353,
|
| 307 |
+
"eval_runtime": 65.1059,
|
| 308 |
+
"eval_samples_per_second": 15.36,
|
| 309 |
+
"eval_steps_per_second": 3.84,
|
| 310 |
+
"step": 3500
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 6.2,
|
| 314 |
+
"grad_norm": 0.4687187671661377,
|
| 315 |
+
"learning_rate": 6.222222222222223e-06,
|
| 316 |
+
"loss": 0.0085,
|
| 317 |
+
"step": 3600
|
| 318 |
+
},
|
| 319 |
+
{
|
| 320 |
+
"epoch": 6.37,
|
| 321 |
+
"grad_norm": 0.696907103061676,
|
| 322 |
+
"learning_rate": 5.777777777777778e-06,
|
| 323 |
+
"loss": 0.0096,
|
| 324 |
+
"step": 3700
|
| 325 |
+
},
|
| 326 |
+
{
|
| 327 |
+
"epoch": 6.54,
|
| 328 |
+
"grad_norm": 0.5151516795158386,
|
| 329 |
+
"learning_rate": 5.333333333333334e-06,
|
| 330 |
+
"loss": 0.009,
|
| 331 |
+
"step": 3800
|
| 332 |
+
},
|
| 333 |
+
{
|
| 334 |
+
"epoch": 6.71,
|
| 335 |
+
"grad_norm": 0.8852140307426453,
|
| 336 |
+
"learning_rate": 4.888888888888889e-06,
|
| 337 |
+
"loss": 0.012,
|
| 338 |
+
"step": 3900
|
| 339 |
+
},
|
| 340 |
+
{
|
| 341 |
+
"epoch": 6.88,
|
| 342 |
+
"grad_norm": 0.8389159440994263,
|
| 343 |
+
"learning_rate": 4.444444444444444e-06,
|
| 344 |
+
"loss": 0.0093,
|
| 345 |
+
"step": 4000
|
| 346 |
+
},
|
| 347 |
+
{
|
| 348 |
+
"epoch": 6.88,
|
| 349 |
+
"eval_loss": 0.3735957741737366,
|
| 350 |
+
"eval_runtime": 65.1742,
|
| 351 |
+
"eval_samples_per_second": 15.343,
|
| 352 |
+
"eval_steps_per_second": 3.836,
|
| 353 |
+
"step": 4000
|
| 354 |
+
},
|
| 355 |
+
{
|
| 356 |
+
"epoch": 7.06,
|
| 357 |
+
"grad_norm": 0.158839613199234,
|
| 358 |
+
"learning_rate": 4.000000000000001e-06,
|
| 359 |
+
"loss": 0.0071,
|
| 360 |
+
"step": 4100
|
| 361 |
+
},
|
| 362 |
+
{
|
| 363 |
+
"epoch": 7.23,
|
| 364 |
+
"grad_norm": 0.4010579586029053,
|
| 365 |
+
"learning_rate": 3.555555555555556e-06,
|
| 366 |
+
"loss": 0.0047,
|
| 367 |
+
"step": 4200
|
| 368 |
+
},
|
| 369 |
+
{
|
| 370 |
+
"epoch": 7.4,
|
| 371 |
+
"grad_norm": 0.16058029234409332,
|
| 372 |
+
"learning_rate": 3.1111111111111116e-06,
|
| 373 |
+
"loss": 0.0037,
|
| 374 |
+
"step": 4300
|
| 375 |
+
},
|
| 376 |
+
{
|
| 377 |
+
"epoch": 7.57,
|
| 378 |
+
"grad_norm": 0.1508949100971222,
|
| 379 |
+
"learning_rate": 2.666666666666667e-06,
|
| 380 |
+
"loss": 0.0039,
|
| 381 |
+
"step": 4400
|
| 382 |
+
},
|
| 383 |
+
{
|
| 384 |
+
"epoch": 7.75,
|
| 385 |
+
"grad_norm": 0.3169790804386139,
|
| 386 |
+
"learning_rate": 2.222222222222222e-06,
|
| 387 |
+
"loss": 0.0037,
|
| 388 |
+
"step": 4500
|
| 389 |
+
},
|
| 390 |
+
{
|
| 391 |
+
"epoch": 7.75,
|
| 392 |
+
"eval_loss": 0.38906845450401306,
|
| 393 |
+
"eval_runtime": 65.0854,
|
| 394 |
+
"eval_samples_per_second": 15.364,
|
| 395 |
+
"eval_steps_per_second": 3.841,
|
| 396 |
+
"step": 4500
|
| 397 |
+
},
|
| 398 |
+
{
|
| 399 |
+
"epoch": 7.92,
|
| 400 |
+
"grad_norm": 0.46604883670806885,
|
| 401 |
+
"learning_rate": 1.777777777777778e-06,
|
| 402 |
+
"loss": 0.0043,
|
| 403 |
+
"step": 4600
|
| 404 |
+
},
|
| 405 |
+
{
|
| 406 |
+
"epoch": 8.09,
|
| 407 |
+
"grad_norm": 0.08523246645927429,
|
| 408 |
+
"learning_rate": 1.3333333333333334e-06,
|
| 409 |
+
"loss": 0.0042,
|
| 410 |
+
"step": 4700
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"epoch": 8.26,
|
| 414 |
+
"grad_norm": 0.2368686944246292,
|
| 415 |
+
"learning_rate": 8.88888888888889e-07,
|
| 416 |
+
"loss": 0.0032,
|
| 417 |
+
"step": 4800
|
| 418 |
+
},
|
| 419 |
+
{
|
| 420 |
+
"epoch": 8.43,
|
| 421 |
+
"grad_norm": 0.11481987684965134,
|
| 422 |
+
"learning_rate": 4.444444444444445e-07,
|
| 423 |
+
"loss": 0.0019,
|
| 424 |
+
"step": 4900
|
| 425 |
+
},
|
| 426 |
+
{
|
| 427 |
+
"epoch": 8.61,
|
| 428 |
+
"grad_norm": 0.13208821415901184,
|
| 429 |
+
"learning_rate": 0.0,
|
| 430 |
+
"loss": 0.002,
|
| 431 |
+
"step": 5000
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 8.61,
|
| 435 |
+
"eval_loss": 0.4305289387702942,
|
| 436 |
+
"eval_runtime": 65.0773,
|
| 437 |
+
"eval_samples_per_second": 15.366,
|
| 438 |
+
"eval_steps_per_second": 3.842,
|
| 439 |
+
"step": 5000
|
| 440 |
}
|
| 441 |
],
|
| 442 |
"logging_steps": 100,
|
|
|
|
| 444 |
"num_input_tokens_seen": 0,
|
| 445 |
"num_train_epochs": 9,
|
| 446 |
"save_steps": 500,
|
| 447 |
+
"total_flos": 1.258569996863275e+18,
|
| 448 |
"train_batch_size": 4,
|
| 449 |
"trial_name": null,
|
| 450 |
"trial_params": null
|