Upload task output 1
Browse files- adapter_config.json +4 -4
- adapter_model.safetensors +1 -1
- loss.txt +1 -1
- trainer_state.json +885 -852
- training_args.bin +1 -1
adapter_config.json
CHANGED
|
@@ -29,13 +29,13 @@
|
|
| 29 |
"rank_pattern": {},
|
| 30 |
"revision": null,
|
| 31 |
"target_modules": [
|
| 32 |
-
"
|
| 33 |
-
"gate_proj",
|
| 34 |
"o_proj",
|
|
|
|
|
|
|
| 35 |
"down_proj",
|
| 36 |
"k_proj",
|
| 37 |
-
"
|
| 38 |
-
"v_proj"
|
| 39 |
],
|
| 40 |
"target_parameters": null,
|
| 41 |
"task_type": "CAUSAL_LM",
|
|
|
|
| 29 |
"rank_pattern": {},
|
| 30 |
"revision": null,
|
| 31 |
"target_modules": [
|
| 32 |
+
"up_proj",
|
|
|
|
| 33 |
"o_proj",
|
| 34 |
+
"v_proj",
|
| 35 |
+
"gate_proj",
|
| 36 |
"down_proj",
|
| 37 |
"k_proj",
|
| 38 |
+
"q_proj"
|
|
|
|
| 39 |
],
|
| 40 |
"target_parameters": null,
|
| 41 |
"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 957942768
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:51dcc60062ae1b1cf80ce9f9cf0b07ba4a42abdffd827bf4c19f5b4c6152a313
|
| 3 |
size 957942768
|
loss.txt
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
|
|
|
|
| 1 |
+
156,-1.1483332872390748
|
trainer_state.json
CHANGED
|
@@ -2,507 +2,507 @@
|
|
| 2 |
"best_global_step": null,
|
| 3 |
"best_metric": null,
|
| 4 |
"best_model_checkpoint": null,
|
| 5 |
-
"epoch": 0.
|
| 6 |
"eval_steps": 500,
|
| 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 |
-
"clip_ratio/high_max": 0.
|
| 14 |
-
"clip_ratio/high_mean": 0.
|
| 15 |
-
"clip_ratio/low_mean": 0.
|
| 16 |
-
"clip_ratio/low_min": 0.
|
| 17 |
-
"clip_ratio/region_mean": 0.
|
| 18 |
"completions/clipped_ratio": 0.0,
|
| 19 |
-
"completions/max_length":
|
| 20 |
-
"completions/max_terminated_length":
|
| 21 |
-
"completions/mean_length":
|
| 22 |
-
"completions/mean_terminated_length":
|
| 23 |
-
"completions/min_length":
|
| 24 |
-
"completions/min_terminated_length":
|
| 25 |
-
"entropy": 0.
|
| 26 |
"epoch": 0.0008,
|
| 27 |
-
"frac_reward_zero_std": 0.
|
| 28 |
-
"grad_norm": 0.
|
| 29 |
-
"kl": 0.
|
| 30 |
"learning_rate": 1.137216e-06,
|
| 31 |
-
"loss":
|
| 32 |
-
"num_tokens":
|
| 33 |
-
"reward": 0.
|
| 34 |
-
"reward_std": 0.
|
| 35 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 36 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 37 |
-
"sampling/importance_sampling_ratio/max": 1.
|
| 38 |
-
"sampling/importance_sampling_ratio/mean":
|
| 39 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 40 |
-
"sampling/sampling_logp_difference/max":
|
| 41 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 42 |
"step": 5,
|
| 43 |
-
"step_time":
|
| 44 |
},
|
| 45 |
{
|
| 46 |
-
"clip_ratio/high_max": 0.
|
| 47 |
-
"clip_ratio/high_mean": 0.
|
| 48 |
-
"clip_ratio/low_mean": 0.
|
| 49 |
"clip_ratio/low_min": 0.0,
|
| 50 |
-
"clip_ratio/region_mean": 0.
|
| 51 |
"completions/clipped_ratio": 0.0,
|
| 52 |
"completions/max_length": 374.0,
|
| 53 |
"completions/max_terminated_length": 374.0,
|
| 54 |
-
"completions/mean_length":
|
| 55 |
-
"completions/mean_terminated_length":
|
| 56 |
-
"completions/min_length":
|
| 57 |
-
"completions/min_terminated_length":
|
| 58 |
-
"entropy": 0.
|
| 59 |
"epoch": 0.0016,
|
| 60 |
-
"frac_reward_zero_std": 0.
|
| 61 |
-
"grad_norm": 0.
|
| 62 |
-
"kl": 0.
|
| 63 |
"learning_rate": 2.5587359999999995e-06,
|
| 64 |
-
"loss":
|
| 65 |
-
"num_tokens":
|
| 66 |
-
"reward": 0.
|
| 67 |
-
"reward_std": 0.
|
| 68 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 69 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 70 |
-
"sampling/importance_sampling_ratio/max": 2.
|
| 71 |
-
"sampling/importance_sampling_ratio/mean":
|
| 72 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 73 |
-
"sampling/sampling_logp_difference/max":
|
| 74 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 75 |
"step": 10,
|
| 76 |
-
"step_time":
|
| 77 |
},
|
| 78 |
{
|
| 79 |
-
"clip_ratio/high_max": 0.
|
| 80 |
-
"clip_ratio/high_mean": 0.
|
| 81 |
-
"clip_ratio/low_mean": 0.
|
| 82 |
-
"clip_ratio/low_min": 0.
|
| 83 |
-
"clip_ratio/region_mean": 0.
|
| 84 |
"completions/clipped_ratio": 0.0,
|
| 85 |
-
"completions/max_length":
|
| 86 |
-
"completions/max_terminated_length":
|
| 87 |
-
"completions/mean_length":
|
| 88 |
-
"completions/mean_terminated_length":
|
| 89 |
-
"completions/min_length":
|
| 90 |
-
"completions/min_terminated_length":
|
| 91 |
-
"entropy": 0.
|
| 92 |
"epoch": 0.0024,
|
| 93 |
-
"frac_reward_zero_std": 0.
|
| 94 |
-
"grad_norm": 0.
|
| 95 |
-
"kl": 0.
|
| 96 |
"learning_rate": 3.9802559999999995e-06,
|
| 97 |
-
"loss":
|
| 98 |
-
"num_tokens":
|
| 99 |
-
"reward": 0.
|
| 100 |
-
"reward_std": 0.
|
| 101 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 102 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 103 |
-
"sampling/importance_sampling_ratio/max": 1.
|
| 104 |
-
"sampling/importance_sampling_ratio/mean": 0.
|
| 105 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 106 |
-
"sampling/sampling_logp_difference/max":
|
| 107 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 108 |
"step": 15,
|
| 109 |
-
"step_time":
|
| 110 |
},
|
| 111 |
{
|
| 112 |
-
"clip_ratio/high_max": 0.
|
| 113 |
-
"clip_ratio/high_mean": 0.
|
| 114 |
-
"clip_ratio/low_mean": 0.
|
| 115 |
-
"clip_ratio/low_min": 0.
|
| 116 |
-
"clip_ratio/region_mean": 0.
|
| 117 |
"completions/clipped_ratio": 0.0,
|
| 118 |
-
"completions/max_length":
|
| 119 |
-
"completions/max_terminated_length":
|
| 120 |
-
"completions/mean_length": 281.
|
| 121 |
-
"completions/mean_terminated_length": 281.
|
| 122 |
"completions/min_length": 212.0,
|
| 123 |
"completions/min_terminated_length": 212.0,
|
| 124 |
-
"entropy": 0.
|
| 125 |
"epoch": 0.0032,
|
| 126 |
-
"frac_reward_zero_std": 0.
|
| 127 |
-
"grad_norm": 0.
|
| 128 |
-
"kl": 0.
|
| 129 |
"learning_rate": 5.401775999999999e-06,
|
| 130 |
-
"loss": -0.
|
| 131 |
-
"num_tokens":
|
| 132 |
-
"reward": 0.
|
| 133 |
-
"reward_std": 0.
|
| 134 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 135 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 136 |
-
"sampling/importance_sampling_ratio/max":
|
| 137 |
-
"sampling/importance_sampling_ratio/mean":
|
| 138 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 139 |
-
"sampling/sampling_logp_difference/max":
|
| 140 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 141 |
"step": 20,
|
| 142 |
-
"step_time":
|
| 143 |
},
|
| 144 |
{
|
| 145 |
-
"clip_ratio/high_max": 0.
|
| 146 |
-
"clip_ratio/high_mean": 0.
|
| 147 |
-
"clip_ratio/low_mean": 0.
|
| 148 |
-
"clip_ratio/low_min": 0.
|
| 149 |
-
"clip_ratio/region_mean": 0.
|
| 150 |
"completions/clipped_ratio": 0.0,
|
| 151 |
-
"completions/max_length": 374.
|
| 152 |
-
"completions/max_terminated_length": 374.
|
| 153 |
-
"completions/mean_length": 300.
|
| 154 |
-
"completions/mean_terminated_length": 300.
|
| 155 |
-
"completions/min_length":
|
| 156 |
-
"completions/min_terminated_length":
|
| 157 |
-
"entropy": 0.
|
| 158 |
"epoch": 0.004,
|
| 159 |
-
"frac_reward_zero_std": 0.
|
| 160 |
-
"grad_norm": 0.
|
| 161 |
-
"kl": 0.
|
| 162 |
"learning_rate": 6.8232959999999994e-06,
|
| 163 |
-
"loss": 0.
|
| 164 |
-
"num_tokens":
|
| 165 |
-
"reward": 0.
|
| 166 |
-
"reward_std": 0.
|
| 167 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 168 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 169 |
-
"sampling/importance_sampling_ratio/max": 1.
|
| 170 |
-
"sampling/importance_sampling_ratio/mean": 0.
|
| 171 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 172 |
-
"sampling/sampling_logp_difference/max":
|
| 173 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 174 |
"step": 25,
|
| 175 |
-
"step_time":
|
| 176 |
},
|
| 177 |
{
|
| 178 |
-
"clip_ratio/high_max": 0.
|
| 179 |
-
"clip_ratio/high_mean": 0.
|
| 180 |
-
"clip_ratio/low_mean": 0.
|
| 181 |
-
"clip_ratio/low_min": 0.
|
| 182 |
-
"clip_ratio/region_mean": 0.
|
| 183 |
"completions/clipped_ratio": 0.0,
|
| 184 |
-
"completions/max_length": 373.
|
| 185 |
-
"completions/max_terminated_length": 373.
|
| 186 |
-
"completions/mean_length": 283.
|
| 187 |
-
"completions/mean_terminated_length": 283.
|
| 188 |
"completions/min_length": 212.0,
|
| 189 |
"completions/min_terminated_length": 212.0,
|
| 190 |
-
"entropy": 0.
|
| 191 |
"epoch": 0.0048,
|
| 192 |
-
"frac_reward_zero_std": 0.
|
| 193 |
-
"grad_norm": 0.
|
| 194 |
-
"kl": 0.
|
| 195 |
"learning_rate": 8.244816e-06,
|
| 196 |
-
"loss": -
|
| 197 |
-
"num_tokens":
|
| 198 |
-
"reward": 0.
|
| 199 |
-
"reward_std": 0.
|
| 200 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 201 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 202 |
-
"sampling/importance_sampling_ratio/max":
|
| 203 |
-
"sampling/importance_sampling_ratio/mean": 0.
|
| 204 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 205 |
-
"sampling/sampling_logp_difference/max":
|
| 206 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 207 |
"step": 30,
|
| 208 |
-
"step_time":
|
| 209 |
},
|
| 210 |
{
|
| 211 |
-
"clip_ratio/high_max": 0.
|
| 212 |
-
"clip_ratio/high_mean": 0.
|
| 213 |
-
"clip_ratio/low_mean": 0.
|
| 214 |
-
"clip_ratio/low_min": 0.
|
| 215 |
-
"clip_ratio/region_mean": 0.
|
| 216 |
"completions/clipped_ratio": 0.0,
|
| 217 |
-
"completions/max_length":
|
| 218 |
-
"completions/max_terminated_length":
|
| 219 |
-
"completions/mean_length": 290.
|
| 220 |
-
"completions/mean_terminated_length": 290.
|
| 221 |
"completions/min_length": 212.0,
|
| 222 |
"completions/min_terminated_length": 212.0,
|
| 223 |
-
"entropy": 0.
|
| 224 |
"epoch": 0.0056,
|
| 225 |
-
"frac_reward_zero_std": 0.
|
| 226 |
-
"grad_norm": 0.
|
| 227 |
-
"kl": 0.
|
| 228 |
"learning_rate": 9.666336e-06,
|
| 229 |
-
"loss": -
|
| 230 |
-
"num_tokens":
|
| 231 |
-
"reward": 0.
|
| 232 |
-
"reward_std": 0.
|
| 233 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 234 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 235 |
-
"sampling/importance_sampling_ratio/max":
|
| 236 |
-
"sampling/importance_sampling_ratio/mean": 0.
|
| 237 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 238 |
-
"sampling/sampling_logp_difference/max":
|
| 239 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 240 |
"step": 35,
|
| 241 |
-
"step_time":
|
| 242 |
},
|
| 243 |
{
|
| 244 |
-
"clip_ratio/high_max": 0.
|
| 245 |
-
"clip_ratio/high_mean": 0.
|
| 246 |
-
"clip_ratio/low_mean": 0.
|
| 247 |
-
"clip_ratio/low_min": 0.
|
| 248 |
-
"clip_ratio/region_mean": 0.
|
| 249 |
"completions/clipped_ratio": 0.0,
|
| 250 |
-
"completions/max_length":
|
| 251 |
-
"completions/max_terminated_length":
|
| 252 |
-
"completions/mean_length":
|
| 253 |
-
"completions/mean_terminated_length":
|
| 254 |
-
"completions/min_length":
|
| 255 |
-
"completions/min_terminated_length":
|
| 256 |
-
"entropy": 0.
|
| 257 |
"epoch": 0.0064,
|
| 258 |
-
"frac_reward_zero_std": 0.
|
| 259 |
-
"grad_norm": 0.
|
| 260 |
-
"kl":
|
| 261 |
"learning_rate": 9.95063915881342e-06,
|
| 262 |
-
"loss": 0.
|
| 263 |
-
"num_tokens":
|
| 264 |
-
"reward": 0.
|
| 265 |
-
"reward_std": 0.
|
| 266 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 267 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 268 |
-
"sampling/importance_sampling_ratio/max":
|
| 269 |
-
"sampling/importance_sampling_ratio/mean":
|
| 270 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 271 |
-
"sampling/sampling_logp_difference/max":
|
| 272 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 273 |
"step": 40,
|
| 274 |
-
"step_time":
|
| 275 |
},
|
| 276 |
{
|
| 277 |
-
"clip_ratio/high_max": 0.
|
| 278 |
-
"clip_ratio/high_mean": 0.
|
| 279 |
-
"clip_ratio/low_mean": 0.
|
| 280 |
"clip_ratio/low_min": 0.0,
|
| 281 |
-
"clip_ratio/region_mean": 0.
|
| 282 |
-
"completions/clipped_ratio": 0.
|
| 283 |
-
"completions/max_length":
|
| 284 |
-
"completions/max_terminated_length": 374.
|
| 285 |
-
"completions/mean_length":
|
| 286 |
-
"completions/mean_terminated_length": 297.
|
| 287 |
"completions/min_length": 212.0,
|
| 288 |
"completions/min_terminated_length": 212.0,
|
| 289 |
-
"entropy": 0.
|
| 290 |
"epoch": 0.0072,
|
| 291 |
-
"frac_reward_zero_std": 0.
|
| 292 |
-
"grad_norm": 0.
|
| 293 |
-
"kl":
|
| 294 |
"learning_rate": 9.950635741493589e-06,
|
| 295 |
-
"loss": 0.
|
| 296 |
-
"num_tokens":
|
| 297 |
-
"reward": 0.
|
| 298 |
-
"reward_std": 0.
|
| 299 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 300 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 301 |
-
"sampling/importance_sampling_ratio/max": 2.
|
| 302 |
-
"sampling/importance_sampling_ratio/mean":
|
| 303 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 304 |
-
"sampling/sampling_logp_difference/max":
|
| 305 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 306 |
"step": 45,
|
| 307 |
-
"step_time":
|
| 308 |
},
|
| 309 |
{
|
| 310 |
-
"clip_ratio/high_max": 0.
|
| 311 |
-
"clip_ratio/high_mean": 0.
|
| 312 |
-
"clip_ratio/low_mean": 0.
|
| 313 |
-
"clip_ratio/low_min": 0.
|
| 314 |
-
"clip_ratio/region_mean": 0.
|
| 315 |
"completions/clipped_ratio": 0.0,
|
| 316 |
-
"completions/max_length":
|
| 317 |
-
"completions/max_terminated_length":
|
| 318 |
-
"completions/mean_length": 299.
|
| 319 |
-
"completions/mean_terminated_length": 299.
|
| 320 |
"completions/min_length": 212.0,
|
| 321 |
"completions/min_terminated_length": 212.0,
|
| 322 |
-
"entropy": 0.
|
| 323 |
"epoch": 0.008,
|
| 324 |
-
"frac_reward_zero_std": 0.
|
| 325 |
-
"grad_norm": 0.
|
| 326 |
-
"kl": 0.
|
| 327 |
"learning_rate": 9.950629695468755e-06,
|
| 328 |
-
"loss": 0.
|
| 329 |
-
"num_tokens":
|
| 330 |
-
"reward": 0.
|
| 331 |
-
"reward_std": 0.
|
| 332 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 333 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 334 |
-
"sampling/importance_sampling_ratio/max":
|
| 335 |
-
"sampling/importance_sampling_ratio/mean": 0.
|
| 336 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 337 |
-
"sampling/sampling_logp_difference/max":
|
| 338 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 339 |
"step": 50,
|
| 340 |
-
"step_time":
|
| 341 |
},
|
| 342 |
{
|
| 343 |
-
"clip_ratio/high_max": 0.
|
| 344 |
-
"clip_ratio/high_mean": 0.
|
| 345 |
-
"clip_ratio/low_mean": 0.
|
| 346 |
-
"clip_ratio/low_min": 0.
|
| 347 |
-
"clip_ratio/region_mean": 0.
|
| 348 |
"completions/clipped_ratio": 0.0,
|
| 349 |
-
"completions/max_length":
|
| 350 |
-
"completions/max_terminated_length":
|
| 351 |
-
"completions/mean_length":
|
| 352 |
-
"completions/mean_terminated_length":
|
| 353 |
-
"completions/min_length":
|
| 354 |
-
"completions/min_terminated_length":
|
| 355 |
-
"entropy": 0.
|
| 356 |
"epoch": 0.0088,
|
| 357 |
-
"frac_reward_zero_std": 0.
|
| 358 |
-
"grad_norm": 0.
|
| 359 |
-
"kl": 0.
|
| 360 |
"learning_rate": 9.950621020743173e-06,
|
| 361 |
-
"loss":
|
| 362 |
-
"num_tokens":
|
| 363 |
-
"reward": 0.
|
| 364 |
-
"reward_std": 0.
|
| 365 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 366 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 367 |
-
"sampling/importance_sampling_ratio/max":
|
| 368 |
-
"sampling/importance_sampling_ratio/mean": 0.
|
| 369 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 370 |
-
"sampling/sampling_logp_difference/max":
|
| 371 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 372 |
"step": 55,
|
| 373 |
-
"step_time":
|
| 374 |
},
|
| 375 |
{
|
| 376 |
-
"clip_ratio/high_max": 0.
|
| 377 |
-
"clip_ratio/high_mean": 0.
|
| 378 |
-
"clip_ratio/low_mean": 0.
|
| 379 |
"clip_ratio/low_min": 0.0,
|
| 380 |
-
"clip_ratio/region_mean": 0.
|
| 381 |
"completions/clipped_ratio": 0.0,
|
| 382 |
-
"completions/max_length":
|
| 383 |
-
"completions/max_terminated_length":
|
| 384 |
-
"completions/mean_length":
|
| 385 |
-
"completions/mean_terminated_length":
|
| 386 |
-
"completions/min_length":
|
| 387 |
-
"completions/min_terminated_length":
|
| 388 |
-
"entropy": 0.
|
| 389 |
"epoch": 0.0096,
|
| 390 |
-
"frac_reward_zero_std": 0.
|
| 391 |
-
"grad_norm": 0.
|
| 392 |
-
"kl":
|
| 393 |
"learning_rate": 9.950609717322956e-06,
|
| 394 |
-
"loss": 0.
|
| 395 |
-
"num_tokens":
|
| 396 |
-
"reward": 0.
|
| 397 |
-
"reward_std": 0.
|
| 398 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 399 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 400 |
-
"sampling/importance_sampling_ratio/max":
|
| 401 |
-
"sampling/importance_sampling_ratio/mean":
|
| 402 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 403 |
-
"sampling/sampling_logp_difference/max":
|
| 404 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 405 |
"step": 60,
|
| 406 |
-
"step_time":
|
| 407 |
},
|
| 408 |
{
|
| 409 |
-
"clip_ratio/high_max": 0.
|
| 410 |
-
"clip_ratio/high_mean": 0.
|
| 411 |
-
"clip_ratio/low_mean": 0.
|
| 412 |
"clip_ratio/low_min": 0.0,
|
| 413 |
-
"clip_ratio/region_mean": 0.
|
| 414 |
"completions/clipped_ratio": 0.0,
|
| 415 |
-
"completions/max_length":
|
| 416 |
-
"completions/max_terminated_length":
|
| 417 |
-
"completions/mean_length":
|
| 418 |
-
"completions/mean_terminated_length":
|
| 419 |
-
"completions/min_length":
|
| 420 |
-
"completions/min_terminated_length":
|
| 421 |
-
"entropy": 0.
|
| 422 |
"epoch": 0.0104,
|
| 423 |
-
"frac_reward_zero_std": 0.
|
| 424 |
-
"grad_norm": 0.
|
| 425 |
-
"kl": 0.
|
| 426 |
"learning_rate": 9.950595785216067e-06,
|
| 427 |
-
"loss": 0.
|
| 428 |
-
"num_tokens":
|
| 429 |
-
"reward": 0.
|
| 430 |
-
"reward_std": 0.
|
| 431 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 432 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 433 |
-
"sampling/importance_sampling_ratio/max":
|
| 434 |
-
"sampling/importance_sampling_ratio/mean": 1.
|
| 435 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 436 |
-
"sampling/sampling_logp_difference/max":
|
| 437 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 438 |
"step": 65,
|
| 439 |
-
"step_time":
|
| 440 |
},
|
| 441 |
{
|
| 442 |
-
"clip_ratio/high_max": 0.
|
| 443 |
-
"clip_ratio/high_mean": 0.
|
| 444 |
-
"clip_ratio/low_mean": 0.
|
| 445 |
"clip_ratio/low_min": 0.0,
|
| 446 |
-
"clip_ratio/region_mean": 0.
|
| 447 |
"completions/clipped_ratio": 0.0,
|
| 448 |
-
"completions/max_length":
|
| 449 |
-
"completions/max_terminated_length":
|
| 450 |
-
"completions/mean_length":
|
| 451 |
-
"completions/mean_terminated_length":
|
| 452 |
-
"completions/min_length":
|
| 453 |
-
"completions/min_terminated_length":
|
| 454 |
-
"entropy": 0.
|
| 455 |
"epoch": 0.0112,
|
| 456 |
-
"frac_reward_zero_std": 0.
|
| 457 |
-
"grad_norm": 0.
|
| 458 |
-
"kl": 1.
|
| 459 |
"learning_rate": 9.950579224432321e-06,
|
| 460 |
-
"loss": 0.
|
| 461 |
-
"num_tokens":
|
| 462 |
-
"reward": 0.
|
| 463 |
-
"reward_std": 0.
|
| 464 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 465 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 466 |
-
"sampling/importance_sampling_ratio/max":
|
| 467 |
-
"sampling/importance_sampling_ratio/mean": 0.
|
| 468 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 469 |
-
"sampling/sampling_logp_difference/max":
|
| 470 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 471 |
"step": 70,
|
| 472 |
-
"step_time": 5.
|
| 473 |
},
|
| 474 |
{
|
| 475 |
"clip_ratio/high_max": 0.0,
|
| 476 |
"clip_ratio/high_mean": 0.0,
|
| 477 |
-
"clip_ratio/low_mean": 0.
|
| 478 |
"clip_ratio/low_min": 0.0,
|
| 479 |
-
"clip_ratio/region_mean": 0.
|
| 480 |
"completions/clipped_ratio": 0.0,
|
| 481 |
-
"completions/max_length":
|
| 482 |
-
"completions/max_terminated_length":
|
| 483 |
-
"completions/mean_length":
|
| 484 |
-
"completions/mean_terminated_length":
|
| 485 |
-
"completions/min_length":
|
| 486 |
-
"completions/min_terminated_length":
|
| 487 |
-
"entropy": 0.
|
| 488 |
"epoch": 0.012,
|
| 489 |
-
"frac_reward_zero_std": 0.
|
| 490 |
-
"grad_norm": 0.
|
| 491 |
-
"kl":
|
| 492 |
"learning_rate": 9.950560034983382e-06,
|
| 493 |
-
"loss": 0.
|
| 494 |
-
"num_tokens":
|
| 495 |
-
"reward": 0.
|
| 496 |
-
"reward_std": 0.
|
| 497 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 498 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 499 |
-
"sampling/importance_sampling_ratio/max":
|
| 500 |
-
"sampling/importance_sampling_ratio/mean": 0.
|
| 501 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 502 |
-
"sampling/sampling_logp_difference/max":
|
| 503 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 504 |
"step": 75,
|
| 505 |
-
"step_time": 5.
|
| 506 |
},
|
| 507 |
{
|
| 508 |
"epoch": 0.012,
|
|
@@ -512,525 +512,525 @@
|
|
| 512 |
"eval_clip_ratio/low_min": 0.0,
|
| 513 |
"eval_clip_ratio/region_mean": 0.0,
|
| 514 |
"eval_completions/clipped_ratio": 0.0,
|
| 515 |
-
"eval_completions/max_length":
|
| 516 |
-
"eval_completions/max_terminated_length":
|
| 517 |
-
"eval_completions/mean_length":
|
| 518 |
-
"eval_completions/mean_terminated_length":
|
| 519 |
-
"eval_completions/min_length":
|
| 520 |
-
"eval_completions/min_terminated_length":
|
| 521 |
-
"eval_entropy": 0.
|
| 522 |
-
"eval_frac_reward_zero_std": 0.
|
| 523 |
-
"eval_kl": 0.
|
| 524 |
-
"eval_loss": -
|
| 525 |
-
"eval_num_tokens":
|
| 526 |
-
"eval_reward": 0.
|
| 527 |
-
"eval_reward_std": 0.
|
| 528 |
-
"eval_rewards/env_goofspiel_reward/mean": 0.
|
| 529 |
-
"eval_rewards/env_goofspiel_reward/std": 0.
|
| 530 |
-
"eval_runtime": 2.
|
| 531 |
-
"eval_samples_per_second":
|
| 532 |
-
"eval_sampling/importance_sampling_ratio/max": 1.
|
| 533 |
-
"eval_sampling/importance_sampling_ratio/mean": 1.
|
| 534 |
-
"eval_sampling/importance_sampling_ratio/min": 0.
|
| 535 |
-
"eval_sampling/sampling_logp_difference/max": 0.
|
| 536 |
-
"eval_sampling/sampling_logp_difference/mean": 0.
|
| 537 |
-
"eval_steps_per_second": 1.
|
| 538 |
"step": 75
|
| 539 |
},
|
| 540 |
{
|
| 541 |
-
"clip_ratio/high_max": 0.
|
| 542 |
-
"clip_ratio/high_mean": 0.
|
| 543 |
-
"clip_ratio/low_mean": 0.
|
| 544 |
-
"clip_ratio/low_min": 0.
|
| 545 |
-
"clip_ratio/region_mean": 0.
|
| 546 |
"completions/clipped_ratio": 0.0,
|
| 547 |
-
"completions/max_length":
|
| 548 |
-
"completions/max_terminated_length":
|
| 549 |
-
"completions/mean_length":
|
| 550 |
-
"completions/mean_terminated_length":
|
| 551 |
-
"completions/min_length":
|
| 552 |
-
"completions/min_terminated_length":
|
| 553 |
-
"entropy": 0.
|
| 554 |
"epoch": 0.0128,
|
| 555 |
-
"frac_reward_zero_std": 0.
|
| 556 |
-
"grad_norm": 0.
|
| 557 |
-
"kl":
|
| 558 |
"learning_rate": 9.95053821688277e-06,
|
| 559 |
-
"loss": 0.
|
| 560 |
-
"num_tokens":
|
| 561 |
-
"reward": 0.
|
| 562 |
-
"reward_std": 0.
|
| 563 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 564 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 565 |
-
"sampling/importance_sampling_ratio/max":
|
| 566 |
-
"sampling/importance_sampling_ratio/mean":
|
| 567 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 568 |
-
"sampling/sampling_logp_difference/max":
|
| 569 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 570 |
"step": 80,
|
| 571 |
-
"step_time": 5.
|
| 572 |
},
|
| 573 |
{
|
| 574 |
-
"clip_ratio/high_max": 0.
|
| 575 |
-
"clip_ratio/high_mean": 0.
|
| 576 |
-
"clip_ratio/low_mean": 0.
|
| 577 |
-
"clip_ratio/low_min": 0.
|
| 578 |
-
"clip_ratio/region_mean": 0.
|
| 579 |
"completions/clipped_ratio": 0.0,
|
| 580 |
-
"completions/max_length":
|
| 581 |
-
"completions/max_terminated_length":
|
| 582 |
-
"completions/mean_length":
|
| 583 |
-
"completions/mean_terminated_length":
|
| 584 |
-
"completions/min_length":
|
| 585 |
-
"completions/min_terminated_length":
|
| 586 |
-
"entropy": 0.
|
| 587 |
"epoch": 0.0136,
|
| 588 |
-
"frac_reward_zero_std": 0.
|
| 589 |
-
"grad_norm": 0.
|
| 590 |
-
"kl":
|
| 591 |
"learning_rate": 9.950513770145857e-06,
|
| 592 |
-
"loss": 0.
|
| 593 |
-
"num_tokens":
|
| 594 |
-
"reward": 0.
|
| 595 |
-
"reward_std": 0.
|
| 596 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 597 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 598 |
-
"sampling/importance_sampling_ratio/max":
|
| 599 |
-
"sampling/importance_sampling_ratio/mean":
|
| 600 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 601 |
-
"sampling/sampling_logp_difference/max":
|
| 602 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 603 |
"step": 85,
|
| 604 |
-
"step_time": 5.
|
| 605 |
},
|
| 606 |
{
|
| 607 |
-
"clip_ratio/high_max": 0.
|
| 608 |
-
"clip_ratio/high_mean": 0.
|
| 609 |
-
"clip_ratio/low_mean": 0.
|
| 610 |
-
"clip_ratio/low_min": 0.
|
| 611 |
-
"clip_ratio/region_mean": 0.
|
| 612 |
"completions/clipped_ratio": 0.0,
|
| 613 |
-
"completions/max_length":
|
| 614 |
-
"completions/max_terminated_length":
|
| 615 |
-
"completions/mean_length":
|
| 616 |
-
"completions/mean_terminated_length":
|
| 617 |
-
"completions/min_length":
|
| 618 |
-
"completions/min_terminated_length":
|
| 619 |
-
"entropy": 0.
|
| 620 |
"epoch": 0.0144,
|
| 621 |
-
"frac_reward_zero_std": 0.
|
| 622 |
-
"grad_norm": 0.
|
| 623 |
-
"kl":
|
| 624 |
"learning_rate": 9.950486694789862e-06,
|
| 625 |
-
"loss": 0.
|
| 626 |
-
"num_tokens":
|
| 627 |
-
"reward": 0.
|
| 628 |
-
"reward_std": 0.
|
| 629 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 630 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 631 |
-
"sampling/importance_sampling_ratio/max": 1.
|
| 632 |
-
"sampling/importance_sampling_ratio/mean":
|
| 633 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 634 |
-
"sampling/sampling_logp_difference/max":
|
| 635 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 636 |
"step": 90,
|
| 637 |
-
"step_time": 5.
|
| 638 |
},
|
| 639 |
{
|
| 640 |
-
"clip_ratio/high_max": 0.
|
| 641 |
-
"clip_ratio/high_mean": 0.
|
| 642 |
-
"clip_ratio/low_mean": 0.
|
| 643 |
-
"clip_ratio/low_min": 0.
|
| 644 |
-
"clip_ratio/region_mean": 0.
|
| 645 |
"completions/clipped_ratio": 0.0,
|
| 646 |
-
"completions/max_length":
|
| 647 |
-
"completions/max_terminated_length":
|
| 648 |
-
"completions/mean_length":
|
| 649 |
-
"completions/mean_terminated_length":
|
| 650 |
-
"completions/min_length":
|
| 651 |
-
"completions/min_terminated_length":
|
| 652 |
-
"entropy": 0.
|
| 653 |
"epoch": 0.0152,
|
| 654 |
-
"frac_reward_zero_std": 0.
|
| 655 |
-
"grad_norm": 0.
|
| 656 |
-
"kl":
|
| 657 |
"learning_rate": 9.95045699083386e-06,
|
| 658 |
-
"loss": 0.
|
| 659 |
-
"num_tokens":
|
| 660 |
-
"reward": 0.
|
| 661 |
-
"reward_std": 0.
|
| 662 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 663 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 664 |
-
"sampling/importance_sampling_ratio/max":
|
| 665 |
-
"sampling/importance_sampling_ratio/mean":
|
| 666 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 667 |
-
"sampling/sampling_logp_difference/max":
|
| 668 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 669 |
"step": 95,
|
| 670 |
-
"step_time": 5.
|
| 671 |
},
|
| 672 |
{
|
| 673 |
"clip_ratio/high_max": 0.0,
|
| 674 |
"clip_ratio/high_mean": 0.0,
|
| 675 |
-
"clip_ratio/low_mean": 0.
|
| 676 |
-
"clip_ratio/low_min": 0.
|
| 677 |
-
"clip_ratio/region_mean": 0.
|
| 678 |
"completions/clipped_ratio": 0.0,
|
| 679 |
-
"completions/max_length":
|
| 680 |
-
"completions/max_terminated_length":
|
| 681 |
-
"completions/mean_length":
|
| 682 |
-
"completions/mean_terminated_length":
|
| 683 |
-
"completions/min_length":
|
| 684 |
-
"completions/min_terminated_length":
|
| 685 |
-
"entropy": 0.
|
| 686 |
"epoch": 0.016,
|
| 687 |
-
"frac_reward_zero_std": 0.
|
| 688 |
-
"grad_norm": 0.
|
| 689 |
-
"kl":
|
| 690 |
"learning_rate": 9.950424658298776e-06,
|
| 691 |
-
"loss": 0.
|
| 692 |
-
"num_tokens":
|
| 693 |
-
"reward": 0.
|
| 694 |
-
"reward_std": 0.
|
| 695 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 696 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 697 |
-
"sampling/importance_sampling_ratio/max": 1.
|
| 698 |
-
"sampling/importance_sampling_ratio/mean": 0.
|
| 699 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 700 |
-
"sampling/sampling_logp_difference/max":
|
| 701 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 702 |
"step": 100,
|
| 703 |
-
"step_time":
|
| 704 |
},
|
| 705 |
{
|
| 706 |
"clip_ratio/high_max": 0.0,
|
| 707 |
"clip_ratio/high_mean": 0.0,
|
| 708 |
-
"clip_ratio/low_mean": 0.
|
| 709 |
"clip_ratio/low_min": 0.0,
|
| 710 |
-
"clip_ratio/region_mean": 0.
|
| 711 |
"completions/clipped_ratio": 0.0,
|
| 712 |
-
"completions/max_length":
|
| 713 |
-
"completions/max_terminated_length":
|
| 714 |
-
"completions/mean_length":
|
| 715 |
-
"completions/mean_terminated_length":
|
| 716 |
-
"completions/min_length":
|
| 717 |
-
"completions/min_terminated_length":
|
| 718 |
-
"entropy": 0.
|
| 719 |
"epoch": 0.0168,
|
| 720 |
-
"frac_reward_zero_std": 0.
|
| 721 |
-
"grad_norm": 0.
|
| 722 |
-
"kl":
|
| 723 |
"learning_rate": 9.950389697207388e-06,
|
| 724 |
-
"loss": 0.
|
| 725 |
-
"num_tokens":
|
| 726 |
-
"reward": 0.
|
| 727 |
-
"reward_std": 0.
|
| 728 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 729 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 730 |
-
"sampling/importance_sampling_ratio/max": 1.
|
| 731 |
-
"sampling/importance_sampling_ratio/mean": 0.
|
| 732 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 733 |
-
"sampling/sampling_logp_difference/max":
|
| 734 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 735 |
"step": 105,
|
| 736 |
-
"step_time":
|
| 737 |
},
|
| 738 |
{
|
| 739 |
-
"clip_ratio/high_max": 0.
|
| 740 |
-
"clip_ratio/high_mean": 0.
|
| 741 |
-
"clip_ratio/low_mean": 0.
|
| 742 |
-
"clip_ratio/low_min": 0.
|
| 743 |
-
"clip_ratio/region_mean": 0.
|
| 744 |
"completions/clipped_ratio": 0.0,
|
| 745 |
-
"completions/max_length":
|
| 746 |
-
"completions/max_terminated_length":
|
| 747 |
-
"completions/mean_length":
|
| 748 |
-
"completions/mean_terminated_length":
|
| 749 |
-
"completions/min_length":
|
| 750 |
-
"completions/min_terminated_length":
|
| 751 |
-
"entropy": 0.
|
| 752 |
"epoch": 0.0176,
|
| 753 |
-
"frac_reward_zero_std": 0.
|
| 754 |
-
"grad_norm": 0.
|
| 755 |
-
"kl":
|
| 756 |
"learning_rate": 9.950352107584324e-06,
|
| 757 |
-
"loss": 0.
|
| 758 |
-
"num_tokens":
|
| 759 |
-
"reward": 0.
|
| 760 |
-
"reward_std": 0.
|
| 761 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 762 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 763 |
-
"sampling/importance_sampling_ratio/max":
|
| 764 |
-
"sampling/importance_sampling_ratio/mean": 0.
|
| 765 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 766 |
-
"sampling/sampling_logp_difference/max":
|
| 767 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 768 |
"step": 110,
|
| 769 |
-
"step_time":
|
| 770 |
},
|
| 771 |
{
|
| 772 |
-
"clip_ratio/high_max": 0.
|
| 773 |
-
"clip_ratio/high_mean": 0.
|
| 774 |
-
"clip_ratio/low_mean": 0.
|
| 775 |
-
"clip_ratio/low_min": 0.
|
| 776 |
-
"clip_ratio/region_mean": 0.
|
| 777 |
"completions/clipped_ratio": 0.0,
|
| 778 |
-
"completions/max_length":
|
| 779 |
-
"completions/max_terminated_length":
|
| 780 |
-
"completions/mean_length":
|
| 781 |
-
"completions/mean_terminated_length":
|
| 782 |
-
"completions/min_length":
|
| 783 |
-
"completions/min_terminated_length":
|
| 784 |
-
"entropy": 0.
|
| 785 |
"epoch": 0.0184,
|
| 786 |
-
"frac_reward_zero_std": 0.
|
| 787 |
-
"grad_norm": 0.
|
| 788 |
-
"kl":
|
| 789 |
"learning_rate": 9.950311889456064e-06,
|
| 790 |
-
"loss": -0.
|
| 791 |
-
"num_tokens":
|
| 792 |
-
"reward":
|
| 793 |
-
"reward_std": 0.
|
| 794 |
-
"rewards/env_goofspiel_reward/mean":
|
| 795 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 796 |
-
"sampling/importance_sampling_ratio/max": 1.
|
| 797 |
-
"sampling/importance_sampling_ratio/mean":
|
| 798 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 799 |
-
"sampling/sampling_logp_difference/max":
|
| 800 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 801 |
"step": 115,
|
| 802 |
-
"step_time":
|
| 803 |
},
|
| 804 |
{
|
| 805 |
"clip_ratio/high_max": 0.0,
|
| 806 |
"clip_ratio/high_mean": 0.0,
|
| 807 |
-
"clip_ratio/low_mean": 0.
|
| 808 |
-
"clip_ratio/low_min": 0.
|
| 809 |
-
"clip_ratio/region_mean": 0.
|
| 810 |
"completions/clipped_ratio": 0.0,
|
| 811 |
-
"completions/max_length":
|
| 812 |
-
"completions/max_terminated_length":
|
| 813 |
-
"completions/mean_length":
|
| 814 |
-
"completions/mean_terminated_length":
|
| 815 |
-
"completions/min_length":
|
| 816 |
-
"completions/min_terminated_length":
|
| 817 |
-
"entropy": 0.
|
| 818 |
"epoch": 0.0192,
|
| 819 |
-
"frac_reward_zero_std": 0.
|
| 820 |
-
"grad_norm": 0.
|
| 821 |
-
"kl":
|
| 822 |
"learning_rate": 9.950269042850943e-06,
|
| 823 |
-
"loss": 0.
|
| 824 |
-
"num_tokens":
|
| 825 |
-
"reward":
|
| 826 |
-
"reward_std": 0.
|
| 827 |
-
"rewards/env_goofspiel_reward/mean":
|
| 828 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 829 |
-
"sampling/importance_sampling_ratio/max": 1.
|
| 830 |
-
"sampling/importance_sampling_ratio/mean": 0.
|
| 831 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 832 |
-
"sampling/sampling_logp_difference/max":
|
| 833 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 834 |
"step": 120,
|
| 835 |
-
"step_time":
|
| 836 |
},
|
| 837 |
{
|
| 838 |
"clip_ratio/high_max": 0.0,
|
| 839 |
"clip_ratio/high_mean": 0.0,
|
| 840 |
-
"clip_ratio/low_mean": 0.
|
| 841 |
"clip_ratio/low_min": 0.0,
|
| 842 |
-
"clip_ratio/region_mean": 0.
|
| 843 |
"completions/clipped_ratio": 0.0,
|
| 844 |
-
"completions/max_length":
|
| 845 |
-
"completions/max_terminated_length":
|
| 846 |
-
"completions/mean_length":
|
| 847 |
-
"completions/mean_terminated_length":
|
| 848 |
-
"completions/min_length":
|
| 849 |
-
"completions/min_terminated_length":
|
| 850 |
-
"entropy": 0.
|
| 851 |
"epoch": 0.02,
|
| 852 |
-
"frac_reward_zero_std": 0.
|
| 853 |
-
"grad_norm": 0.
|
| 854 |
-
"kl":
|
| 855 |
"learning_rate": 9.95022356779914e-06,
|
| 856 |
-
"loss": -
|
| 857 |
-
"num_tokens":
|
| 858 |
-
"reward": 0.
|
| 859 |
-
"reward_std": 0.
|
| 860 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 861 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 862 |
-
"sampling/importance_sampling_ratio/max": 1.
|
| 863 |
-
"sampling/importance_sampling_ratio/mean":
|
| 864 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 865 |
-
"sampling/sampling_logp_difference/max":
|
| 866 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 867 |
"step": 125,
|
| 868 |
-
"step_time":
|
| 869 |
},
|
| 870 |
{
|
| 871 |
-
"clip_ratio/high_max": 0.
|
| 872 |
-
"clip_ratio/high_mean": 0.
|
| 873 |
-
"clip_ratio/low_mean": 0.
|
| 874 |
-
"clip_ratio/low_min": 0.
|
| 875 |
-
"clip_ratio/region_mean": 0.
|
| 876 |
"completions/clipped_ratio": 0.0,
|
| 877 |
-
"completions/max_length":
|
| 878 |
-
"completions/max_terminated_length":
|
| 879 |
-
"completions/mean_length":
|
| 880 |
-
"completions/mean_terminated_length":
|
| 881 |
-
"completions/min_length":
|
| 882 |
-
"completions/min_terminated_length":
|
| 883 |
-
"entropy": 0.
|
| 884 |
"epoch": 0.0208,
|
| 885 |
-
"frac_reward_zero_std": 0.
|
| 886 |
-
"grad_norm": 0.
|
| 887 |
-
"kl":
|
| 888 |
"learning_rate": 9.950175464332696e-06,
|
| 889 |
-
"loss":
|
| 890 |
-
"num_tokens":
|
| 891 |
-
"reward": 0.
|
| 892 |
-
"reward_std": 0.
|
| 893 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 894 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 895 |
-
"sampling/importance_sampling_ratio/max":
|
| 896 |
-
"sampling/importance_sampling_ratio/mean":
|
| 897 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 898 |
-
"sampling/sampling_logp_difference/max":
|
| 899 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 900 |
"step": 130,
|
| 901 |
-
"step_time":
|
| 902 |
},
|
| 903 |
{
|
| 904 |
"clip_ratio/high_max": 0.0,
|
| 905 |
"clip_ratio/high_mean": 0.0,
|
| 906 |
-
"clip_ratio/low_mean": 0.
|
| 907 |
-
"clip_ratio/low_min": 0.
|
| 908 |
-
"clip_ratio/region_mean": 0.
|
| 909 |
"completions/clipped_ratio": 0.0,
|
| 910 |
-
"completions/max_length":
|
| 911 |
-
"completions/max_terminated_length":
|
| 912 |
-
"completions/mean_length":
|
| 913 |
-
"completions/mean_terminated_length":
|
| 914 |
-
"completions/min_length":
|
| 915 |
-
"completions/min_terminated_length":
|
| 916 |
-
"entropy": 0.
|
| 917 |
"epoch": 0.0216,
|
| 918 |
-
"frac_reward_zero_std": 0.
|
| 919 |
-
"grad_norm": 0.
|
| 920 |
-
"kl":
|
| 921 |
"learning_rate": 9.950124732485496e-06,
|
| 922 |
-
"loss": -0.
|
| 923 |
-
"num_tokens":
|
| 924 |
-
"reward": 0.
|
| 925 |
-
"reward_std": 0.
|
| 926 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 927 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 928 |
-
"sampling/importance_sampling_ratio/max":
|
| 929 |
-
"sampling/importance_sampling_ratio/mean":
|
| 930 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 931 |
-
"sampling/sampling_logp_difference/max":
|
| 932 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 933 |
"step": 135,
|
| 934 |
-
"step_time":
|
| 935 |
},
|
| 936 |
{
|
| 937 |
-
"clip_ratio/high_max": 0.
|
| 938 |
-
"clip_ratio/high_mean": 0.
|
| 939 |
-
"clip_ratio/low_mean": 0.
|
| 940 |
"clip_ratio/low_min": 0.0,
|
| 941 |
-
"clip_ratio/region_mean": 0.
|
| 942 |
"completions/clipped_ratio": 0.0,
|
| 943 |
-
"completions/max_length":
|
| 944 |
-
"completions/max_terminated_length":
|
| 945 |
-
"completions/mean_length":
|
| 946 |
-
"completions/mean_terminated_length":
|
| 947 |
-
"completions/min_length":
|
| 948 |
-
"completions/min_terminated_length":
|
| 949 |
-
"entropy": 0.
|
| 950 |
"epoch": 0.0224,
|
| 951 |
-
"frac_reward_zero_std": 0.
|
| 952 |
-
"grad_norm": 0.
|
| 953 |
-
"kl":
|
| 954 |
"learning_rate": 9.95007137229328e-06,
|
| 955 |
-
"loss":
|
| 956 |
-
"num_tokens":
|
| 957 |
-
"reward": 0.
|
| 958 |
-
"reward_std": 0.
|
| 959 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 960 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 961 |
-
"sampling/importance_sampling_ratio/max":
|
| 962 |
-
"sampling/importance_sampling_ratio/mean": 0.
|
| 963 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 964 |
-
"sampling/sampling_logp_difference/max":
|
| 965 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 966 |
"step": 140,
|
| 967 |
-
"step_time":
|
| 968 |
},
|
| 969 |
{
|
| 970 |
-
"clip_ratio/high_max": 0.
|
| 971 |
-
"clip_ratio/high_mean": 0.
|
| 972 |
-
"clip_ratio/low_mean": 0.
|
| 973 |
-
"clip_ratio/low_min": 0.
|
| 974 |
-
"clip_ratio/region_mean": 0.
|
| 975 |
"completions/clipped_ratio": 0.0,
|
| 976 |
-
"completions/max_length":
|
| 977 |
-
"completions/max_terminated_length":
|
| 978 |
-
"completions/mean_length":
|
| 979 |
-
"completions/mean_terminated_length":
|
| 980 |
-
"completions/min_length":
|
| 981 |
-
"completions/min_terminated_length":
|
| 982 |
-
"entropy": 0.
|
| 983 |
"epoch": 0.0232,
|
| 984 |
-
"frac_reward_zero_std": 0.
|
| 985 |
-
"grad_norm": 0.
|
| 986 |
-
"kl":
|
| 987 |
"learning_rate": 9.950015383793636e-06,
|
| 988 |
-
"loss": -0.
|
| 989 |
-
"num_tokens":
|
| 990 |
-
"reward": 0.
|
| 991 |
-
"reward_std": 0.
|
| 992 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 993 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 994 |
-
"sampling/importance_sampling_ratio/max":
|
| 995 |
-
"sampling/importance_sampling_ratio/mean":
|
| 996 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 997 |
-
"sampling/sampling_logp_difference/max":
|
| 998 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 999 |
"step": 145,
|
| 1000 |
-
"step_time":
|
| 1001 |
},
|
| 1002 |
{
|
| 1003 |
-
"clip_ratio/high_max": 0.
|
| 1004 |
-
"clip_ratio/high_mean": 0.
|
| 1005 |
-
"clip_ratio/low_mean": 0.
|
| 1006 |
"clip_ratio/low_min": 0.0,
|
| 1007 |
-
"clip_ratio/region_mean": 0.
|
| 1008 |
"completions/clipped_ratio": 0.0,
|
| 1009 |
-
"completions/max_length":
|
| 1010 |
-
"completions/max_terminated_length":
|
| 1011 |
-
"completions/mean_length":
|
| 1012 |
-
"completions/mean_terminated_length":
|
| 1013 |
-
"completions/min_length":
|
| 1014 |
-
"completions/min_terminated_length":
|
| 1015 |
-
"entropy": 0.
|
| 1016 |
"epoch": 0.024,
|
| 1017 |
-
"frac_reward_zero_std": 0.
|
| 1018 |
-
"grad_norm": 0.
|
| 1019 |
-
"kl":
|
| 1020 |
"learning_rate": 9.949956767026006e-06,
|
| 1021 |
-
"loss": 0.
|
| 1022 |
-
"num_tokens":
|
| 1023 |
-
"reward": 0.
|
| 1024 |
-
"reward_std": 0.
|
| 1025 |
-
"rewards/env_goofspiel_reward/mean": 0.
|
| 1026 |
-
"rewards/env_goofspiel_reward/std": 0.
|
| 1027 |
-
"sampling/importance_sampling_ratio/max":
|
| 1028 |
-
"sampling/importance_sampling_ratio/mean": 1.
|
| 1029 |
-
"sampling/importance_sampling_ratio/min": 0.
|
| 1030 |
-
"sampling/sampling_logp_difference/max":
|
| 1031 |
-
"sampling/sampling_logp_difference/mean": 0.
|
| 1032 |
"step": 150,
|
| 1033 |
-
"step_time":
|
| 1034 |
},
|
| 1035 |
{
|
| 1036 |
"epoch": 0.024,
|
|
@@ -1040,68 +1040,101 @@
|
|
| 1040 |
"eval_clip_ratio/low_min": 0.0,
|
| 1041 |
"eval_clip_ratio/region_mean": 0.0,
|
| 1042 |
"eval_completions/clipped_ratio": 0.0,
|
| 1043 |
-
"eval_completions/max_length":
|
| 1044 |
-
"eval_completions/max_terminated_length":
|
| 1045 |
-
"eval_completions/mean_length":
|
| 1046 |
-
"eval_completions/mean_terminated_length":
|
| 1047 |
-
"eval_completions/min_length":
|
| 1048 |
-
"eval_completions/min_terminated_length":
|
| 1049 |
-
"eval_entropy": 0.
|
| 1050 |
-
"eval_frac_reward_zero_std": 0.
|
| 1051 |
-
"eval_kl":
|
| 1052 |
-
"eval_loss": -0.
|
| 1053 |
-
"eval_num_tokens":
|
| 1054 |
-
"eval_reward":
|
| 1055 |
-
"eval_reward_std": 0.
|
| 1056 |
-
"eval_rewards/env_goofspiel_reward/mean":
|
| 1057 |
-
"eval_rewards/env_goofspiel_reward/std": 0.
|
| 1058 |
-
"eval_runtime":
|
| 1059 |
-
"eval_samples_per_second":
|
| 1060 |
-
"eval_sampling/importance_sampling_ratio/max": 1.
|
| 1061 |
-
"eval_sampling/importance_sampling_ratio/mean":
|
| 1062 |
-
"eval_sampling/importance_sampling_ratio/min": 0.
|
| 1063 |
-
"eval_sampling/sampling_logp_difference/max": 0.
|
| 1064 |
-
"eval_sampling/sampling_logp_difference/mean": 0.
|
| 1065 |
-
"eval_steps_per_second": 0.
|
| 1066 |
"step": 150
|
| 1067 |
},
|
| 1068 |
{
|
| 1069 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1070 |
"eval_clip_ratio/high_max": 0.0,
|
| 1071 |
"eval_clip_ratio/high_mean": 0.0,
|
| 1072 |
"eval_clip_ratio/low_mean": 0.0,
|
| 1073 |
"eval_clip_ratio/low_min": 0.0,
|
| 1074 |
"eval_clip_ratio/region_mean": 0.0,
|
| 1075 |
"eval_completions/clipped_ratio": 0.0,
|
| 1076 |
-
"eval_completions/max_length":
|
| 1077 |
-
"eval_completions/max_terminated_length":
|
| 1078 |
-
"eval_completions/mean_length":
|
| 1079 |
-
"eval_completions/mean_terminated_length":
|
| 1080 |
-
"eval_completions/min_length":
|
| 1081 |
-
"eval_completions/min_terminated_length":
|
| 1082 |
-
"eval_entropy": 0.
|
| 1083 |
-
"eval_frac_reward_zero_std":
|
| 1084 |
-
"eval_kl":
|
| 1085 |
-
"eval_loss":
|
| 1086 |
-
"eval_num_tokens":
|
| 1087 |
-
"eval_reward":
|
| 1088 |
-
"eval_reward_std": 0.
|
| 1089 |
-
"eval_rewards/env_goofspiel_reward/mean":
|
| 1090 |
-
"eval_rewards/env_goofspiel_reward/std": 0.
|
| 1091 |
-
"eval_runtime":
|
| 1092 |
-
"eval_samples_per_second":
|
| 1093 |
-
"eval_sampling/importance_sampling_ratio/max": 1.
|
| 1094 |
-
"eval_sampling/importance_sampling_ratio/mean": 1.
|
| 1095 |
-
"eval_sampling/importance_sampling_ratio/min": 0.
|
| 1096 |
-
"eval_sampling/sampling_logp_difference/max": 0.
|
| 1097 |
-
"eval_sampling/sampling_logp_difference/mean": 0.
|
| 1098 |
-
"eval_steps_per_second": 0.
|
| 1099 |
-
"step":
|
| 1100 |
}
|
| 1101 |
],
|
| 1102 |
"logging_steps": 5,
|
| 1103 |
"max_steps": 18750,
|
| 1104 |
-
"num_input_tokens_seen":
|
| 1105 |
"num_train_epochs": 3,
|
| 1106 |
"save_steps": 500,
|
| 1107 |
"stateful_callbacks": {
|
|
@@ -1117,7 +1150,7 @@
|
|
| 1117 |
}
|
| 1118 |
},
|
| 1119 |
"total_flos": 0.0,
|
| 1120 |
-
"train_batch_size":
|
| 1121 |
"trial_name": null,
|
| 1122 |
"trial_params": null
|
| 1123 |
}
|
|
|
|
| 2 |
"best_global_step": null,
|
| 3 |
"best_metric": null,
|
| 4 |
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.02496,
|
| 6 |
"eval_steps": 500,
|
| 7 |
+
"global_step": 156,
|
| 8 |
"is_hyper_param_search": false,
|
| 9 |
"is_local_process_zero": true,
|
| 10 |
"is_world_process_zero": true,
|
| 11 |
"log_history": [
|
| 12 |
{
|
| 13 |
+
"clip_ratio/high_max": 0.01656746044754982,
|
| 14 |
+
"clip_ratio/high_mean": 0.00828373022377491,
|
| 15 |
+
"clip_ratio/low_mean": 0.009885329194366932,
|
| 16 |
+
"clip_ratio/low_min": 0.002380952425301075,
|
| 17 |
+
"clip_ratio/region_mean": 0.018169059325009583,
|
| 18 |
"completions/clipped_ratio": 0.0,
|
| 19 |
+
"completions/max_length": 373.8,
|
| 20 |
+
"completions/max_terminated_length": 373.8,
|
| 21 |
+
"completions/mean_length": 296.71875,
|
| 22 |
+
"completions/mean_terminated_length": 296.71875,
|
| 23 |
+
"completions/min_length": 206.8,
|
| 24 |
+
"completions/min_terminated_length": 206.8,
|
| 25 |
+
"entropy": 0.35859955847263336,
|
| 26 |
"epoch": 0.0008,
|
| 27 |
+
"frac_reward_zero_std": 0.475,
|
| 28 |
+
"grad_norm": 0.11806713044643402,
|
| 29 |
+
"kl": 0.010710475360974669,
|
| 30 |
"learning_rate": 1.137216e-06,
|
| 31 |
+
"loss": 0.00025553270243108275,
|
| 32 |
+
"num_tokens": 136400.0,
|
| 33 |
+
"reward": 0.2433750033378601,
|
| 34 |
+
"reward_std": 0.22857226729393004,
|
| 35 |
+
"rewards/env_goofspiel_reward/mean": 0.2433750033378601,
|
| 36 |
+
"rewards/env_goofspiel_reward/std": 0.3573280215263367,
|
| 37 |
+
"sampling/importance_sampling_ratio/max": 1.642977237701416,
|
| 38 |
+
"sampling/importance_sampling_ratio/mean": 0.9299223780632019,
|
| 39 |
+
"sampling/importance_sampling_ratio/min": 0.2294576272368431,
|
| 40 |
+
"sampling/sampling_logp_difference/max": 1.239228868484497,
|
| 41 |
+
"sampling/sampling_logp_difference/mean": 0.08511904180049897,
|
| 42 |
"step": 5,
|
| 43 |
+
"step_time": 5.990162861399949
|
| 44 |
},
|
| 45 |
{
|
| 46 |
+
"clip_ratio/high_max": 0.03551879096776247,
|
| 47 |
+
"clip_ratio/high_mean": 0.019321895483881236,
|
| 48 |
+
"clip_ratio/low_mean": 0.010304330103099345,
|
| 49 |
"clip_ratio/low_min": 0.0,
|
| 50 |
+
"clip_ratio/region_mean": 0.02962622568011284,
|
| 51 |
"completions/clipped_ratio": 0.0,
|
| 52 |
"completions/max_length": 374.0,
|
| 53 |
"completions/max_terminated_length": 374.0,
|
| 54 |
+
"completions/mean_length": 291.0,
|
| 55 |
+
"completions/mean_terminated_length": 291.0,
|
| 56 |
+
"completions/min_length": 199.6,
|
| 57 |
+
"completions/min_terminated_length": 199.6,
|
| 58 |
+
"entropy": 0.3628120839595795,
|
| 59 |
"epoch": 0.0016,
|
| 60 |
+
"frac_reward_zero_std": 0.425,
|
| 61 |
+
"grad_norm": 0.11497774720191956,
|
| 62 |
+
"kl": 0.014784016623161733,
|
| 63 |
"learning_rate": 2.5587359999999995e-06,
|
| 64 |
+
"loss": 0.0005161482840776443,
|
| 65 |
+
"num_tokens": 271519.0,
|
| 66 |
+
"reward": 0.3373125076293945,
|
| 67 |
+
"reward_std": 0.28646664023399354,
|
| 68 |
+
"rewards/env_goofspiel_reward/mean": 0.3373125076293945,
|
| 69 |
+
"rewards/env_goofspiel_reward/std": 0.39762375950813295,
|
| 70 |
+
"sampling/importance_sampling_ratio/max": 2.0244646072387695,
|
| 71 |
+
"sampling/importance_sampling_ratio/mean": 0.95240398645401,
|
| 72 |
+
"sampling/importance_sampling_ratio/min": 0.3043131291866302,
|
| 73 |
+
"sampling/sampling_logp_difference/max": 1.4355120182037353,
|
| 74 |
+
"sampling/sampling_logp_difference/mean": 0.08179195821285248,
|
| 75 |
"step": 10,
|
| 76 |
+
"step_time": 5.4302939728002455
|
| 77 |
},
|
| 78 |
{
|
| 79 |
+
"clip_ratio/high_max": 0.020830108411610128,
|
| 80 |
+
"clip_ratio/high_mean": 0.01188564244657755,
|
| 81 |
+
"clip_ratio/low_mean": 0.015411754604429007,
|
| 82 |
+
"clip_ratio/low_min": 0.004805492050945759,
|
| 83 |
+
"clip_ratio/region_mean": 0.0272973969578743,
|
| 84 |
"completions/clipped_ratio": 0.0,
|
| 85 |
+
"completions/max_length": 374.0,
|
| 86 |
+
"completions/max_terminated_length": 374.0,
|
| 87 |
+
"completions/mean_length": 282.2875,
|
| 88 |
+
"completions/mean_terminated_length": 282.2875,
|
| 89 |
+
"completions/min_length": 189.6,
|
| 90 |
+
"completions/min_terminated_length": 189.6,
|
| 91 |
+
"entropy": 0.35702232643961906,
|
| 92 |
"epoch": 0.0024,
|
| 93 |
+
"frac_reward_zero_std": 0.4875,
|
| 94 |
+
"grad_norm": 0.15851116180419922,
|
| 95 |
+
"kl": 0.03802298142109066,
|
| 96 |
"learning_rate": 3.9802559999999995e-06,
|
| 97 |
+
"loss": -8.745418745093048e-05,
|
| 98 |
+
"num_tokens": 404021.0,
|
| 99 |
+
"reward": 0.30337501466274264,
|
| 100 |
+
"reward_std": 0.2497854709625244,
|
| 101 |
+
"rewards/env_goofspiel_reward/mean": 0.30337501466274264,
|
| 102 |
+
"rewards/env_goofspiel_reward/std": 0.37030403017997743,
|
| 103 |
+
"sampling/importance_sampling_ratio/max": 1.8557178020477294,
|
| 104 |
+
"sampling/importance_sampling_ratio/mean": 0.9347150444984436,
|
| 105 |
+
"sampling/importance_sampling_ratio/min": 0.2671460062265396,
|
| 106 |
+
"sampling/sampling_logp_difference/max": 1.3799697399139403,
|
| 107 |
+
"sampling/sampling_logp_difference/mean": 0.08675528764724731,
|
| 108 |
"step": 15,
|
| 109 |
+
"step_time": 5.612129885600188
|
| 110 |
},
|
| 111 |
{
|
| 112 |
+
"clip_ratio/high_max": 0.017504085041582584,
|
| 113 |
+
"clip_ratio/high_mean": 0.010222630854696036,
|
| 114 |
+
"clip_ratio/low_mean": 0.016076818853616715,
|
| 115 |
+
"clip_ratio/low_min": 0.005441176518797874,
|
| 116 |
+
"clip_ratio/region_mean": 0.02629944970831275,
|
| 117 |
"completions/clipped_ratio": 0.0,
|
| 118 |
+
"completions/max_length": 373.8,
|
| 119 |
+
"completions/max_terminated_length": 373.8,
|
| 120 |
+
"completions/mean_length": 281.4375,
|
| 121 |
+
"completions/mean_terminated_length": 281.4375,
|
| 122 |
"completions/min_length": 212.0,
|
| 123 |
"completions/min_terminated_length": 212.0,
|
| 124 |
+
"entropy": 0.33125333562493325,
|
| 125 |
"epoch": 0.0032,
|
| 126 |
+
"frac_reward_zero_std": 0.4875,
|
| 127 |
+
"grad_norm": 0.3056468069553375,
|
| 128 |
+
"kl": 0.2386752042453736,
|
| 129 |
"learning_rate": 5.401775999999999e-06,
|
| 130 |
+
"loss": -0.00010562442475929857,
|
| 131 |
+
"num_tokens": 536239.0,
|
| 132 |
+
"reward": 0.33362500965595243,
|
| 133 |
+
"reward_std": 0.22821871638298036,
|
| 134 |
+
"rewards/env_goofspiel_reward/mean": 0.33362500965595243,
|
| 135 |
+
"rewards/env_goofspiel_reward/std": 0.3822357296943665,
|
| 136 |
+
"sampling/importance_sampling_ratio/max": 2.044223356246948,
|
| 137 |
+
"sampling/importance_sampling_ratio/mean": 0.9417648911476135,
|
| 138 |
+
"sampling/importance_sampling_ratio/min": 0.1664988785982132,
|
| 139 |
+
"sampling/sampling_logp_difference/max": 1.9717662334442139,
|
| 140 |
+
"sampling/sampling_logp_difference/mean": 0.09318876564502716,
|
| 141 |
"step": 20,
|
| 142 |
+
"step_time": 5.200572347599882
|
| 143 |
},
|
| 144 |
{
|
| 145 |
+
"clip_ratio/high_max": 0.014910130761563778,
|
| 146 |
+
"clip_ratio/high_mean": 0.007455065380781889,
|
| 147 |
+
"clip_ratio/low_mean": 0.014673632569611073,
|
| 148 |
+
"clip_ratio/low_min": 0.005902777798473835,
|
| 149 |
+
"clip_ratio/region_mean": 0.02212869795039296,
|
| 150 |
"completions/clipped_ratio": 0.0,
|
| 151 |
+
"completions/max_length": 374.4,
|
| 152 |
+
"completions/max_terminated_length": 374.4,
|
| 153 |
+
"completions/mean_length": 300.20625,
|
| 154 |
+
"completions/mean_terminated_length": 300.20625,
|
| 155 |
+
"completions/min_length": 214.2,
|
| 156 |
+
"completions/min_terminated_length": 214.2,
|
| 157 |
+
"entropy": 0.3311316817998886,
|
| 158 |
"epoch": 0.004,
|
| 159 |
+
"frac_reward_zero_std": 0.5,
|
| 160 |
+
"grad_norm": 0.24050310254096985,
|
| 161 |
+
"kl": 0.2951220686081797,
|
| 162 |
"learning_rate": 6.8232959999999994e-06,
|
| 163 |
+
"loss": 0.00011087653692811727,
|
| 164 |
+
"num_tokens": 674105.0,
|
| 165 |
+
"reward": 0.3523750066757202,
|
| 166 |
+
"reward_std": 0.22291541397571563,
|
| 167 |
+
"rewards/env_goofspiel_reward/mean": 0.3523750066757202,
|
| 168 |
+
"rewards/env_goofspiel_reward/std": 0.39020044207572935,
|
| 169 |
+
"sampling/importance_sampling_ratio/max": 1.9415390253067017,
|
| 170 |
+
"sampling/importance_sampling_ratio/mean": 0.9240148186683654,
|
| 171 |
+
"sampling/importance_sampling_ratio/min": 0.07138497838750482,
|
| 172 |
+
"sampling/sampling_logp_difference/max": 2.9348490715026854,
|
| 173 |
+
"sampling/sampling_logp_difference/mean": 0.10956742316484451,
|
| 174 |
"step": 25,
|
| 175 |
+
"step_time": 5.527079869400405
|
| 176 |
},
|
| 177 |
{
|
| 178 |
+
"clip_ratio/high_max": 0.012152777798473836,
|
| 179 |
+
"clip_ratio/high_mean": 0.006076388899236918,
|
| 180 |
+
"clip_ratio/low_mean": 0.01880923202261329,
|
| 181 |
+
"clip_ratio/low_min": 0.005718954280018807,
|
| 182 |
+
"clip_ratio/region_mean": 0.024885620921850204,
|
| 183 |
"completions/clipped_ratio": 0.0,
|
| 184 |
+
"completions/max_length": 373.4,
|
| 185 |
+
"completions/max_terminated_length": 373.4,
|
| 186 |
+
"completions/mean_length": 283.375,
|
| 187 |
+
"completions/mean_terminated_length": 283.375,
|
| 188 |
"completions/min_length": 212.0,
|
| 189 |
"completions/min_terminated_length": 212.0,
|
| 190 |
+
"entropy": 0.32351960614323616,
|
| 191 |
"epoch": 0.0048,
|
| 192 |
+
"frac_reward_zero_std": 0.4125,
|
| 193 |
+
"grad_norm": 0.12574604153633118,
|
| 194 |
+
"kl": 0.5364798322319985,
|
| 195 |
"learning_rate": 8.244816e-06,
|
| 196 |
+
"loss": -0.0002830292796716094,
|
| 197 |
+
"num_tokens": 805805.0,
|
| 198 |
+
"reward": 0.4162500202655792,
|
| 199 |
+
"reward_std": 0.30228814482688904,
|
| 200 |
+
"rewards/env_goofspiel_reward/mean": 0.4162500202655792,
|
| 201 |
+
"rewards/env_goofspiel_reward/std": 0.4285269021987915,
|
| 202 |
+
"sampling/importance_sampling_ratio/max": 2.1214988470077514,
|
| 203 |
+
"sampling/importance_sampling_ratio/mean": 0.9363813877105713,
|
| 204 |
+
"sampling/importance_sampling_ratio/min": 0.08070152997970581,
|
| 205 |
+
"sampling/sampling_logp_difference/max": 1.7284237623214722,
|
| 206 |
+
"sampling/sampling_logp_difference/mean": 0.1223350703716278,
|
| 207 |
"step": 30,
|
| 208 |
+
"step_time": 5.215961435999452
|
| 209 |
},
|
| 210 |
{
|
| 211 |
+
"clip_ratio/high_max": 0.011638931930065155,
|
| 212 |
+
"clip_ratio/high_mean": 0.0058194659650325775,
|
| 213 |
+
"clip_ratio/low_mean": 0.01746001038700342,
|
| 214 |
+
"clip_ratio/low_min": 0.003125,
|
| 215 |
+
"clip_ratio/region_mean": 0.023279476352035998,
|
| 216 |
"completions/clipped_ratio": 0.0,
|
| 217 |
+
"completions/max_length": 366.0,
|
| 218 |
+
"completions/max_terminated_length": 366.0,
|
| 219 |
+
"completions/mean_length": 290.51875,
|
| 220 |
+
"completions/mean_terminated_length": 290.51875,
|
| 221 |
"completions/min_length": 212.0,
|
| 222 |
"completions/min_terminated_length": 212.0,
|
| 223 |
+
"entropy": 0.33921139910817144,
|
| 224 |
"epoch": 0.0056,
|
| 225 |
+
"frac_reward_zero_std": 0.3875,
|
| 226 |
+
"grad_norm": 0.14579056203365326,
|
| 227 |
+
"kl": 0.4404270384460688,
|
| 228 |
"learning_rate": 9.666336e-06,
|
| 229 |
+
"loss": -0.0004368364345282316,
|
| 230 |
+
"num_tokens": 940358.0,
|
| 231 |
+
"reward": 0.42375001311302185,
|
| 232 |
+
"reward_std": 0.3022881507873535,
|
| 233 |
+
"rewards/env_goofspiel_reward/mean": 0.42375001311302185,
|
| 234 |
+
"rewards/env_goofspiel_reward/std": 0.4120412886142731,
|
| 235 |
+
"sampling/importance_sampling_ratio/max": 2.12521378993988,
|
| 236 |
+
"sampling/importance_sampling_ratio/mean": 0.896282148361206,
|
| 237 |
+
"sampling/importance_sampling_ratio/min": 0.10518757700920105,
|
| 238 |
+
"sampling/sampling_logp_difference/max": 1.934821891784668,
|
| 239 |
+
"sampling/sampling_logp_difference/mean": 0.14520370960235596,
|
| 240 |
"step": 35,
|
| 241 |
+
"step_time": 5.421745918600391
|
| 242 |
},
|
| 243 |
{
|
| 244 |
+
"clip_ratio/high_max": 0.010828081332147122,
|
| 245 |
+
"clip_ratio/high_mean": 0.005414040666073561,
|
| 246 |
+
"clip_ratio/low_mean": 0.02190126050263643,
|
| 247 |
+
"clip_ratio/low_min": 0.008496732078492641,
|
| 248 |
+
"clip_ratio/region_mean": 0.02731530126184225,
|
| 249 |
"completions/clipped_ratio": 0.0,
|
| 250 |
+
"completions/max_length": 374.4,
|
| 251 |
+
"completions/max_terminated_length": 374.4,
|
| 252 |
+
"completions/mean_length": 288.0625,
|
| 253 |
+
"completions/mean_terminated_length": 288.0625,
|
| 254 |
+
"completions/min_length": 196.8,
|
| 255 |
+
"completions/min_terminated_length": 196.8,
|
| 256 |
+
"entropy": 0.23345143683254718,
|
| 257 |
"epoch": 0.0064,
|
| 258 |
+
"frac_reward_zero_std": 0.4,
|
| 259 |
+
"grad_norm": 0.07758636772632599,
|
| 260 |
+
"kl": 2.4490169264376163,
|
| 261 |
"learning_rate": 9.95063915881342e-06,
|
| 262 |
+
"loss": -0.00021622611675411462,
|
| 263 |
+
"num_tokens": 1075088.0,
|
| 264 |
+
"reward": 0.49831252098083495,
|
| 265 |
+
"reward_std": 0.24987386167049408,
|
| 266 |
+
"rewards/env_goofspiel_reward/mean": 0.49831252098083495,
|
| 267 |
+
"rewards/env_goofspiel_reward/std": 0.39479002356529236,
|
| 268 |
+
"sampling/importance_sampling_ratio/max": 2.2193925380706787,
|
| 269 |
+
"sampling/importance_sampling_ratio/mean": 0.8783411741256714,
|
| 270 |
+
"sampling/importance_sampling_ratio/min": 0.005415213014930487,
|
| 271 |
+
"sampling/sampling_logp_difference/max": 3.553822565078735,
|
| 272 |
+
"sampling/sampling_logp_difference/mean": 0.16359376013278962,
|
| 273 |
"step": 40,
|
| 274 |
+
"step_time": 5.113493552800537
|
| 275 |
},
|
| 276 |
{
|
| 277 |
+
"clip_ratio/high_max": 0.011822755448520184,
|
| 278 |
+
"clip_ratio/high_mean": 0.005911377724260092,
|
| 279 |
+
"clip_ratio/low_mean": 0.015286377724260091,
|
| 280 |
"clip_ratio/low_min": 0.0,
|
| 281 |
+
"clip_ratio/region_mean": 0.021197755355387926,
|
| 282 |
+
"completions/clipped_ratio": 0.0,
|
| 283 |
+
"completions/max_length": 374.2,
|
| 284 |
+
"completions/max_terminated_length": 374.2,
|
| 285 |
+
"completions/mean_length": 297.9,
|
| 286 |
+
"completions/mean_terminated_length": 297.9,
|
| 287 |
"completions/min_length": 212.0,
|
| 288 |
"completions/min_terminated_length": 212.0,
|
| 289 |
+
"entropy": 0.2887581422924995,
|
| 290 |
"epoch": 0.0072,
|
| 291 |
+
"frac_reward_zero_std": 0.4375,
|
| 292 |
+
"grad_norm": 0.07674326002597809,
|
| 293 |
+
"kl": 1.0048415780067443,
|
| 294 |
"learning_rate": 9.950635741493589e-06,
|
| 295 |
+
"loss": -0.0004248165525496006,
|
| 296 |
+
"num_tokens": 1212402.0,
|
| 297 |
+
"reward": 0.5099375128746033,
|
| 298 |
+
"reward_std": 0.24404022693634034,
|
| 299 |
+
"rewards/env_goofspiel_reward/mean": 0.5099375128746033,
|
| 300 |
+
"rewards/env_goofspiel_reward/std": 0.3780271291732788,
|
| 301 |
+
"sampling/importance_sampling_ratio/max": 2.1818215370178224,
|
| 302 |
+
"sampling/importance_sampling_ratio/mean": 0.8968270421028137,
|
| 303 |
+
"sampling/importance_sampling_ratio/min": 0.040483005344867706,
|
| 304 |
+
"sampling/sampling_logp_difference/max": 2.9256459712982177,
|
| 305 |
+
"sampling/sampling_logp_difference/mean": 0.14963440895080565,
|
| 306 |
"step": 45,
|
| 307 |
+
"step_time": 5.549473219600077
|
| 308 |
},
|
| 309 |
{
|
| 310 |
+
"clip_ratio/high_max": 0.014889705926179886,
|
| 311 |
+
"clip_ratio/high_mean": 0.007444852963089943,
|
| 312 |
+
"clip_ratio/low_mean": 0.009685134887695313,
|
| 313 |
+
"clip_ratio/low_min": 0.005409356765449047,
|
| 314 |
+
"clip_ratio/region_mean": 0.017129987850785254,
|
| 315 |
"completions/clipped_ratio": 0.0,
|
| 316 |
+
"completions/max_length": 374.2,
|
| 317 |
+
"completions/max_terminated_length": 374.2,
|
| 318 |
+
"completions/mean_length": 299.5625,
|
| 319 |
+
"completions/mean_terminated_length": 299.5625,
|
| 320 |
"completions/min_length": 212.0,
|
| 321 |
"completions/min_terminated_length": 212.0,
|
| 322 |
+
"entropy": 0.35890122279524805,
|
| 323 |
"epoch": 0.008,
|
| 324 |
+
"frac_reward_zero_std": 0.3375,
|
| 325 |
+
"grad_norm": 0.0733598917722702,
|
| 326 |
+
"kl": 0.8497510462999344,
|
| 327 |
"learning_rate": 9.950629695468755e-06,
|
| 328 |
+
"loss": -0.0006576518062502146,
|
| 329 |
+
"num_tokens": 1349581.0,
|
| 330 |
+
"reward": 0.5286250114440918,
|
| 331 |
+
"reward_std": 0.33428473472595216,
|
| 332 |
+
"rewards/env_goofspiel_reward/mean": 0.5286250114440918,
|
| 333 |
+
"rewards/env_goofspiel_reward/std": 0.4310214161872864,
|
| 334 |
+
"sampling/importance_sampling_ratio/max": 2.438156485557556,
|
| 335 |
+
"sampling/importance_sampling_ratio/mean": 0.9875588178634643,
|
| 336 |
+
"sampling/importance_sampling_ratio/min": 0.08052640929818153,
|
| 337 |
+
"sampling/sampling_logp_difference/max": 1.630429244041443,
|
| 338 |
+
"sampling/sampling_logp_difference/mean": 0.12703385949134827,
|
| 339 |
"step": 50,
|
| 340 |
+
"step_time": 5.318245979599669
|
| 341 |
},
|
| 342 |
{
|
| 343 |
+
"clip_ratio/high_max": 0.008750000037252903,
|
| 344 |
+
"clip_ratio/high_mean": 0.005937500018626451,
|
| 345 |
+
"clip_ratio/low_mean": 0.008613153640180827,
|
| 346 |
+
"clip_ratio/low_min": 0.002777777798473835,
|
| 347 |
+
"clip_ratio/region_mean": 0.014550653658807278,
|
| 348 |
"completions/clipped_ratio": 0.0,
|
| 349 |
+
"completions/max_length": 373.6,
|
| 350 |
+
"completions/max_terminated_length": 373.6,
|
| 351 |
+
"completions/mean_length": 282.6625,
|
| 352 |
+
"completions/mean_terminated_length": 282.6625,
|
| 353 |
+
"completions/min_length": 201.6,
|
| 354 |
+
"completions/min_terminated_length": 201.6,
|
| 355 |
+
"entropy": 0.324828689545393,
|
| 356 |
"epoch": 0.0088,
|
| 357 |
+
"frac_reward_zero_std": 0.375,
|
| 358 |
+
"grad_norm": 0.07007352262735367,
|
| 359 |
+
"kl": 0.7210939504206181,
|
| 360 |
"learning_rate": 9.950621020743173e-06,
|
| 361 |
+
"loss": -0.0003687262535095215,
|
| 362 |
+
"num_tokens": 1481810.0,
|
| 363 |
+
"reward": 0.4908750057220459,
|
| 364 |
+
"reward_std": 0.29203510880470274,
|
| 365 |
+
"rewards/env_goofspiel_reward/mean": 0.4908750057220459,
|
| 366 |
+
"rewards/env_goofspiel_reward/std": 0.38116695880889895,
|
| 367 |
+
"sampling/importance_sampling_ratio/max": 2.387629532814026,
|
| 368 |
+
"sampling/importance_sampling_ratio/mean": 0.9571901321411133,
|
| 369 |
+
"sampling/importance_sampling_ratio/min": 0.1296325594186783,
|
| 370 |
+
"sampling/sampling_logp_difference/max": 1.4196331977844239,
|
| 371 |
+
"sampling/sampling_logp_difference/mean": 0.11004247218370437,
|
| 372 |
"step": 55,
|
| 373 |
+
"step_time": 5.2706628203999575
|
| 374 |
},
|
| 375 |
{
|
| 376 |
+
"clip_ratio/high_max": 0.008534356765449048,
|
| 377 |
+
"clip_ratio/high_mean": 0.005656067188829184,
|
| 378 |
+
"clip_ratio/low_mean": 0.009407982788980007,
|
| 379 |
"clip_ratio/low_min": 0.0,
|
| 380 |
+
"clip_ratio/region_mean": 0.01506404997780919,
|
| 381 |
"completions/clipped_ratio": 0.0,
|
| 382 |
+
"completions/max_length": 374.2,
|
| 383 |
+
"completions/max_terminated_length": 374.2,
|
| 384 |
+
"completions/mean_length": 293.0875,
|
| 385 |
+
"completions/mean_terminated_length": 293.0875,
|
| 386 |
+
"completions/min_length": 213.8,
|
| 387 |
+
"completions/min_terminated_length": 213.8,
|
| 388 |
+
"entropy": 0.33867746219038963,
|
| 389 |
"epoch": 0.0096,
|
| 390 |
+
"frac_reward_zero_std": 0.4,
|
| 391 |
+
"grad_norm": 0.07561016827821732,
|
| 392 |
+
"kl": 1.2898097231984138,
|
| 393 |
"learning_rate": 9.950609717322956e-06,
|
| 394 |
+
"loss": -0.0007428391836583614,
|
| 395 |
+
"num_tokens": 1617445.0,
|
| 396 |
+
"reward": 0.5697500050067902,
|
| 397 |
+
"reward_std": 0.26551859080791473,
|
| 398 |
+
"rewards/env_goofspiel_reward/mean": 0.5697500050067902,
|
| 399 |
+
"rewards/env_goofspiel_reward/std": 0.3711147367954254,
|
| 400 |
+
"sampling/importance_sampling_ratio/max": 2.2963364124298096,
|
| 401 |
+
"sampling/importance_sampling_ratio/mean": 0.976230013370514,
|
| 402 |
+
"sampling/importance_sampling_ratio/min": 0.07447343468666076,
|
| 403 |
+
"sampling/sampling_logp_difference/max": 1.4920601844787598,
|
| 404 |
+
"sampling/sampling_logp_difference/mean": 0.1161011055111885,
|
| 405 |
"step": 60,
|
| 406 |
+
"step_time": 5.45707285199951
|
| 407 |
},
|
| 408 |
{
|
| 409 |
+
"clip_ratio/high_max": 0.006066176481544972,
|
| 410 |
+
"clip_ratio/high_mean": 0.003033088240772486,
|
| 411 |
+
"clip_ratio/low_mean": 0.00883374186232686,
|
| 412 |
"clip_ratio/low_min": 0.0,
|
| 413 |
+
"clip_ratio/region_mean": 0.011866830103099347,
|
| 414 |
"completions/clipped_ratio": 0.0,
|
| 415 |
+
"completions/max_length": 373.8,
|
| 416 |
+
"completions/max_terminated_length": 373.8,
|
| 417 |
+
"completions/mean_length": 273.0875,
|
| 418 |
+
"completions/mean_terminated_length": 273.0875,
|
| 419 |
+
"completions/min_length": 199.6,
|
| 420 |
+
"completions/min_terminated_length": 199.6,
|
| 421 |
+
"entropy": 0.2858745433390141,
|
| 422 |
"epoch": 0.0104,
|
| 423 |
+
"frac_reward_zero_std": 0.4625,
|
| 424 |
+
"grad_norm": 0.05500200390815735,
|
| 425 |
+
"kl": 0.8528277602046728,
|
| 426 |
"learning_rate": 9.950595785216067e-06,
|
| 427 |
+
"loss": -0.0008720592595636845,
|
| 428 |
+
"num_tokens": 1746271.0,
|
| 429 |
+
"reward": 0.6673750162124634,
|
| 430 |
+
"reward_std": 0.2547352105379105,
|
| 431 |
+
"rewards/env_goofspiel_reward/mean": 0.6673750162124634,
|
| 432 |
+
"rewards/env_goofspiel_reward/std": 0.35124587416648867,
|
| 433 |
+
"sampling/importance_sampling_ratio/max": 2.0680487632751463,
|
| 434 |
+
"sampling/importance_sampling_ratio/mean": 1.0014554977416992,
|
| 435 |
+
"sampling/importance_sampling_ratio/min": 0.10332663245499134,
|
| 436 |
+
"sampling/sampling_logp_difference/max": 2.259426403045654,
|
| 437 |
+
"sampling/sampling_logp_difference/mean": 0.10407083481550217,
|
| 438 |
"step": 65,
|
| 439 |
+
"step_time": 5.311621308399845
|
| 440 |
},
|
| 441 |
{
|
| 442 |
+
"clip_ratio/high_max": 0.01180555559694767,
|
| 443 |
+
"clip_ratio/high_mean": 0.005902777798473835,
|
| 444 |
+
"clip_ratio/low_mean": 0.005365742836147547,
|
| 445 |
"clip_ratio/low_min": 0.0,
|
| 446 |
+
"clip_ratio/region_mean": 0.011268520634621382,
|
| 447 |
"completions/clipped_ratio": 0.0,
|
| 448 |
+
"completions/max_length": 374.2,
|
| 449 |
+
"completions/max_terminated_length": 374.2,
|
| 450 |
+
"completions/mean_length": 279.90625,
|
| 451 |
+
"completions/mean_terminated_length": 279.90625,
|
| 452 |
+
"completions/min_length": 206.8,
|
| 453 |
+
"completions/min_terminated_length": 206.8,
|
| 454 |
+
"entropy": 0.24370604529976844,
|
| 455 |
"epoch": 0.0112,
|
| 456 |
+
"frac_reward_zero_std": 0.475,
|
| 457 |
+
"grad_norm": 0.055401790887117386,
|
| 458 |
+
"kl": 1.1588016405701638,
|
| 459 |
"learning_rate": 9.950579224432321e-06,
|
| 460 |
+
"loss": -0.0012944268994033337,
|
| 461 |
+
"num_tokens": 1878445.0,
|
| 462 |
+
"reward": 0.7311875104904175,
|
| 463 |
+
"reward_std": 0.24934353530406952,
|
| 464 |
+
"rewards/env_goofspiel_reward/mean": 0.7311875104904175,
|
| 465 |
+
"rewards/env_goofspiel_reward/std": 0.3471538543701172,
|
| 466 |
+
"sampling/importance_sampling_ratio/max": 2.0970391750335695,
|
| 467 |
+
"sampling/importance_sampling_ratio/mean": 0.9700004577636718,
|
| 468 |
+
"sampling/importance_sampling_ratio/min": 0.16687118411064147,
|
| 469 |
+
"sampling/sampling_logp_difference/max": 1.9643285274505615,
|
| 470 |
+
"sampling/sampling_logp_difference/mean": 0.10032870918512345,
|
| 471 |
"step": 70,
|
| 472 |
+
"step_time": 5.481714394999836
|
| 473 |
},
|
| 474 |
{
|
| 475 |
"clip_ratio/high_max": 0.0,
|
| 476 |
"clip_ratio/high_mean": 0.0,
|
| 477 |
+
"clip_ratio/low_mean": 0.013843201845884324,
|
| 478 |
"clip_ratio/low_min": 0.0,
|
| 479 |
+
"clip_ratio/region_mean": 0.013843201845884324,
|
| 480 |
"completions/clipped_ratio": 0.0,
|
| 481 |
+
"completions/max_length": 374.4,
|
| 482 |
+
"completions/max_terminated_length": 374.4,
|
| 483 |
+
"completions/mean_length": 291.925,
|
| 484 |
+
"completions/mean_terminated_length": 291.925,
|
| 485 |
+
"completions/min_length": 199.6,
|
| 486 |
+
"completions/min_terminated_length": 199.6,
|
| 487 |
+
"entropy": 0.2260144092142582,
|
| 488 |
"epoch": 0.012,
|
| 489 |
+
"frac_reward_zero_std": 0.475,
|
| 490 |
+
"grad_norm": 0.023790325969457626,
|
| 491 |
+
"kl": 4.869625084102154,
|
| 492 |
"learning_rate": 9.950560034983382e-06,
|
| 493 |
+
"loss": -0.000660108495503664,
|
| 494 |
+
"num_tokens": 2013416.0,
|
| 495 |
+
"reward": 0.7499375104904175,
|
| 496 |
+
"reward_std": 0.26525343060493467,
|
| 497 |
+
"rewards/env_goofspiel_reward/mean": 0.7499375104904175,
|
| 498 |
+
"rewards/env_goofspiel_reward/std": 0.3587073266506195,
|
| 499 |
+
"sampling/importance_sampling_ratio/max": 2.49392306804657,
|
| 500 |
+
"sampling/importance_sampling_ratio/mean": 0.9810920596122742,
|
| 501 |
+
"sampling/importance_sampling_ratio/min": 0.07269524885341525,
|
| 502 |
+
"sampling/sampling_logp_difference/max": 2.6866398811340333,
|
| 503 |
+
"sampling/sampling_logp_difference/mean": 0.09891045838594437,
|
| 504 |
"step": 75,
|
| 505 |
+
"step_time": 5.191597324399481
|
| 506 |
},
|
| 507 |
{
|
| 508 |
"epoch": 0.012,
|
|
|
|
| 512 |
"eval_clip_ratio/low_min": 0.0,
|
| 513 |
"eval_clip_ratio/region_mean": 0.0,
|
| 514 |
"eval_completions/clipped_ratio": 0.0,
|
| 515 |
+
"eval_completions/max_length": 290.0,
|
| 516 |
+
"eval_completions/max_terminated_length": 290.0,
|
| 517 |
+
"eval_completions/mean_length": 269.7,
|
| 518 |
+
"eval_completions/mean_terminated_length": 269.7,
|
| 519 |
+
"eval_completions/min_length": 249.4,
|
| 520 |
+
"eval_completions/min_terminated_length": 249.4,
|
| 521 |
+
"eval_entropy": 0.2245341122150421,
|
| 522 |
+
"eval_frac_reward_zero_std": 0.4,
|
| 523 |
+
"eval_kl": 0.6459955453872681,
|
| 524 |
+
"eval_loss": -0.00020105746807530522,
|
| 525 |
+
"eval_num_tokens": 2013416.0,
|
| 526 |
+
"eval_reward": 0.9000000357627869,
|
| 527 |
+
"eval_reward_std": 0.2545584440231323,
|
| 528 |
+
"eval_rewards/env_goofspiel_reward/mean": 0.9000000357627869,
|
| 529 |
+
"eval_rewards/env_goofspiel_reward/std": 0.32784610986709595,
|
| 530 |
+
"eval_runtime": 2.4336,
|
| 531 |
+
"eval_samples_per_second": 4.109,
|
| 532 |
+
"eval_sampling/importance_sampling_ratio/max": 1.3313236951828002,
|
| 533 |
+
"eval_sampling/importance_sampling_ratio/mean": 1.0307578563690185,
|
| 534 |
+
"eval_sampling/importance_sampling_ratio/min": 0.6750483691692353,
|
| 535 |
+
"eval_sampling/sampling_logp_difference/max": 0.6310659945011139,
|
| 536 |
+
"eval_sampling/sampling_logp_difference/mean": 0.06487421616911888,
|
| 537 |
+
"eval_steps_per_second": 1.233,
|
| 538 |
"step": 75
|
| 539 |
},
|
| 540 |
{
|
| 541 |
+
"clip_ratio/high_max": 0.005441176518797874,
|
| 542 |
+
"clip_ratio/high_mean": 0.002720588259398937,
|
| 543 |
+
"clip_ratio/low_mean": 0.008915441203862428,
|
| 544 |
+
"clip_ratio/low_min": 0.0029411764815449716,
|
| 545 |
+
"clip_ratio/region_mean": 0.011636029463261366,
|
| 546 |
"completions/clipped_ratio": 0.0,
|
| 547 |
+
"completions/max_length": 374.4,
|
| 548 |
+
"completions/max_terminated_length": 374.4,
|
| 549 |
+
"completions/mean_length": 283.0125,
|
| 550 |
+
"completions/mean_terminated_length": 283.0125,
|
| 551 |
+
"completions/min_length": 206.8,
|
| 552 |
+
"completions/min_terminated_length": 206.8,
|
| 553 |
+
"entropy": 0.2153526909649372,
|
| 554 |
"epoch": 0.0128,
|
| 555 |
+
"frac_reward_zero_std": 0.4625,
|
| 556 |
+
"grad_norm": 0.07206813246011734,
|
| 557 |
+
"kl": 1.6877675473690033,
|
| 558 |
"learning_rate": 9.95053821688277e-06,
|
| 559 |
+
"loss": -0.0009287594817578793,
|
| 560 |
+
"num_tokens": 2145819.0,
|
| 561 |
+
"reward": 0.7724375128746033,
|
| 562 |
+
"reward_std": 0.24404022693634034,
|
| 563 |
+
"rewards/env_goofspiel_reward/mean": 0.7724375128746033,
|
| 564 |
+
"rewards/env_goofspiel_reward/std": 0.34512641429901125,
|
| 565 |
+
"sampling/importance_sampling_ratio/max": 2.1818724155426024,
|
| 566 |
+
"sampling/importance_sampling_ratio/mean": 0.9245784163475037,
|
| 567 |
+
"sampling/importance_sampling_ratio/min": 0.13156846463680266,
|
| 568 |
+
"sampling/sampling_logp_difference/max": 1.658913779258728,
|
| 569 |
+
"sampling/sampling_logp_difference/mean": 0.10279093384742737,
|
| 570 |
"step": 80,
|
| 571 |
+
"step_time": 5.441123167599471
|
| 572 |
},
|
| 573 |
{
|
| 574 |
+
"clip_ratio/high_max": 0.0029411764815449716,
|
| 575 |
+
"clip_ratio/high_mean": 0.0014705882407724858,
|
| 576 |
+
"clip_ratio/low_mean": 0.011678629741072654,
|
| 577 |
+
"clip_ratio/low_min": 0.0029411764815449716,
|
| 578 |
+
"clip_ratio/region_mean": 0.013149217981845141,
|
| 579 |
"completions/clipped_ratio": 0.0,
|
| 580 |
+
"completions/max_length": 373.6,
|
| 581 |
+
"completions/max_terminated_length": 373.6,
|
| 582 |
+
"completions/mean_length": 287.66875,
|
| 583 |
+
"completions/mean_terminated_length": 287.66875,
|
| 584 |
+
"completions/min_length": 201.6,
|
| 585 |
+
"completions/min_terminated_length": 201.6,
|
| 586 |
+
"entropy": 0.2192211326211691,
|
| 587 |
"epoch": 0.0136,
|
| 588 |
+
"frac_reward_zero_std": 0.3625,
|
| 589 |
+
"grad_norm": 0.049152813851833344,
|
| 590 |
+
"kl": 1.2483771443367004,
|
| 591 |
"learning_rate": 9.950513770145857e-06,
|
| 592 |
+
"loss": -0.0017359975725412368,
|
| 593 |
+
"num_tokens": 2279605.0,
|
| 594 |
+
"reward": 0.8173750400543213,
|
| 595 |
+
"reward_std": 0.3077682375907898,
|
| 596 |
+
"rewards/env_goofspiel_reward/mean": 0.8173750400543213,
|
| 597 |
+
"rewards/env_goofspiel_reward/std": 0.4139078199863434,
|
| 598 |
+
"sampling/importance_sampling_ratio/max": 2.272643804550171,
|
| 599 |
+
"sampling/importance_sampling_ratio/mean": 0.9856367349624634,
|
| 600 |
+
"sampling/importance_sampling_ratio/min": 0.07546308934688568,
|
| 601 |
+
"sampling/sampling_logp_difference/max": 1.7427294969558715,
|
| 602 |
+
"sampling/sampling_logp_difference/mean": 0.118245729804039,
|
| 603 |
"step": 85,
|
| 604 |
+
"step_time": 5.293755201000204
|
| 605 |
},
|
| 606 |
{
|
| 607 |
+
"clip_ratio/high_max": 0.005263157933950424,
|
| 608 |
+
"clip_ratio/high_mean": 0.002631578966975212,
|
| 609 |
+
"clip_ratio/low_mean": 0.010241443105041981,
|
| 610 |
+
"clip_ratio/low_min": 0.002631578966975212,
|
| 611 |
+
"clip_ratio/region_mean": 0.012873022072017192,
|
| 612 |
"completions/clipped_ratio": 0.0,
|
| 613 |
+
"completions/max_length": 374.4,
|
| 614 |
+
"completions/max_terminated_length": 374.4,
|
| 615 |
+
"completions/mean_length": 297.4875,
|
| 616 |
+
"completions/mean_terminated_length": 297.4875,
|
| 617 |
+
"completions/min_length": 207.0,
|
| 618 |
+
"completions/min_terminated_length": 207.0,
|
| 619 |
+
"entropy": 0.19745248332619666,
|
| 620 |
"epoch": 0.0144,
|
| 621 |
+
"frac_reward_zero_std": 0.375,
|
| 622 |
+
"grad_norm": 0.10243905335664749,
|
| 623 |
+
"kl": 1.8384233698248864,
|
| 624 |
"learning_rate": 9.950486694789862e-06,
|
| 625 |
+
"loss": -0.0016141118481755256,
|
| 626 |
+
"num_tokens": 2415904.0,
|
| 627 |
+
"reward": 0.8171875357627869,
|
| 628 |
+
"reward_std": 0.28682020902633665,
|
| 629 |
+
"rewards/env_goofspiel_reward/mean": 0.8171875357627869,
|
| 630 |
+
"rewards/env_goofspiel_reward/std": 0.4000221848487854,
|
| 631 |
+
"sampling/importance_sampling_ratio/max": 1.893965482711792,
|
| 632 |
+
"sampling/importance_sampling_ratio/mean": 0.9034451127052308,
|
| 633 |
+
"sampling/importance_sampling_ratio/min": 0.10746023580431938,
|
| 634 |
+
"sampling/sampling_logp_difference/max": 2.5204232692718507,
|
| 635 |
+
"sampling/sampling_logp_difference/mean": 0.11648025661706925,
|
| 636 |
"step": 90,
|
| 637 |
+
"step_time": 5.317576567400283
|
| 638 |
},
|
| 639 |
{
|
| 640 |
+
"clip_ratio/high_max": 0.001666666753590107,
|
| 641 |
+
"clip_ratio/high_mean": 0.0008333333767950535,
|
| 642 |
+
"clip_ratio/low_mean": 0.011362766660749913,
|
| 643 |
+
"clip_ratio/low_min": 0.002631578966975212,
|
| 644 |
+
"clip_ratio/region_mean": 0.012196100037544965,
|
| 645 |
"completions/clipped_ratio": 0.0,
|
| 646 |
+
"completions/max_length": 374.6,
|
| 647 |
+
"completions/max_terminated_length": 374.6,
|
| 648 |
+
"completions/mean_length": 282.925,
|
| 649 |
+
"completions/mean_terminated_length": 282.925,
|
| 650 |
+
"completions/min_length": 199.6,
|
| 651 |
+
"completions/min_terminated_length": 199.6,
|
| 652 |
+
"entropy": 0.14891982078552246,
|
| 653 |
"epoch": 0.0152,
|
| 654 |
+
"frac_reward_zero_std": 0.4625,
|
| 655 |
+
"grad_norm": 0.022762339562177658,
|
| 656 |
+
"kl": 1.1775232434272767,
|
| 657 |
"learning_rate": 9.95045699083386e-06,
|
| 658 |
+
"loss": -0.00033199070021510123,
|
| 659 |
+
"num_tokens": 2548327.0,
|
| 660 |
+
"reward": 0.8961250424385071,
|
| 661 |
+
"reward_std": 0.28125171959400175,
|
| 662 |
+
"rewards/env_goofspiel_reward/mean": 0.8961250424385071,
|
| 663 |
+
"rewards/env_goofspiel_reward/std": 0.39407379031181333,
|
| 664 |
+
"sampling/importance_sampling_ratio/max": 2.3227984428405763,
|
| 665 |
+
"sampling/importance_sampling_ratio/mean": 0.9894420146942139,
|
| 666 |
+
"sampling/importance_sampling_ratio/min": 0.1872235879302025,
|
| 667 |
+
"sampling/sampling_logp_difference/max": 1.8188178777694701,
|
| 668 |
+
"sampling/sampling_logp_difference/mean": 0.0839442029595375,
|
| 669 |
"step": 95,
|
| 670 |
+
"step_time": 5.388430412399975
|
| 671 |
},
|
| 672 |
{
|
| 673 |
"clip_ratio/high_max": 0.0,
|
| 674 |
"clip_ratio/high_mean": 0.0,
|
| 675 |
+
"clip_ratio/low_mean": 0.013613153528422118,
|
| 676 |
+
"clip_ratio/low_min": 0.0029411764815449716,
|
| 677 |
+
"clip_ratio/region_mean": 0.013613153528422118,
|
| 678 |
"completions/clipped_ratio": 0.0,
|
| 679 |
+
"completions/max_length": 374.2,
|
| 680 |
+
"completions/max_terminated_length": 374.2,
|
| 681 |
+
"completions/mean_length": 285.875,
|
| 682 |
+
"completions/mean_terminated_length": 285.875,
|
| 683 |
+
"completions/min_length": 218.8,
|
| 684 |
+
"completions/min_terminated_length": 218.8,
|
| 685 |
+
"entropy": 0.16100836284458636,
|
| 686 |
"epoch": 0.016,
|
| 687 |
+
"frac_reward_zero_std": 0.5,
|
| 688 |
+
"grad_norm": 0.03195883333683014,
|
| 689 |
+
"kl": 1.7148397326469422,
|
| 690 |
"learning_rate": 9.950424658298776e-06,
|
| 691 |
+
"loss": -0.00081931222230196,
|
| 692 |
+
"num_tokens": 2682046.0,
|
| 693 |
+
"reward": 0.9037500023841858,
|
| 694 |
+
"reward_std": 0.23864853382110596,
|
| 695 |
+
"rewards/env_goofspiel_reward/mean": 0.9037500023841858,
|
| 696 |
+
"rewards/env_goofspiel_reward/std": 0.37243106961250305,
|
| 697 |
+
"sampling/importance_sampling_ratio/max": 1.8023416996002197,
|
| 698 |
+
"sampling/importance_sampling_ratio/mean": 0.8842753887176513,
|
| 699 |
+
"sampling/importance_sampling_ratio/min": 0.10478578060865402,
|
| 700 |
+
"sampling/sampling_logp_difference/max": 1.7476533651351929,
|
| 701 |
+
"sampling/sampling_logp_difference/mean": 0.09434186965227127,
|
| 702 |
"step": 100,
|
| 703 |
+
"step_time": 5.138518711399956
|
| 704 |
},
|
| 705 |
{
|
| 706 |
"clip_ratio/high_max": 0.0,
|
| 707 |
"clip_ratio/high_mean": 0.0,
|
| 708 |
+
"clip_ratio/low_mean": 0.0045955882407724856,
|
| 709 |
"clip_ratio/low_min": 0.0,
|
| 710 |
+
"clip_ratio/region_mean": 0.0045955882407724856,
|
| 711 |
"completions/clipped_ratio": 0.0,
|
| 712 |
+
"completions/max_length": 374.4,
|
| 713 |
+
"completions/max_terminated_length": 374.4,
|
| 714 |
+
"completions/mean_length": 288.55,
|
| 715 |
+
"completions/mean_terminated_length": 288.55,
|
| 716 |
+
"completions/min_length": 212.0,
|
| 717 |
+
"completions/min_terminated_length": 212.0,
|
| 718 |
+
"entropy": 0.15357921570539473,
|
| 719 |
"epoch": 0.0168,
|
| 720 |
+
"frac_reward_zero_std": 0.475,
|
| 721 |
+
"grad_norm": 0.02917584963142872,
|
| 722 |
+
"kl": 1.2142755538225174,
|
| 723 |
"learning_rate": 9.950389697207388e-06,
|
| 724 |
+
"loss": -0.0009133240208029747,
|
| 725 |
+
"num_tokens": 2815770.0,
|
| 726 |
+
"reward": 0.9600000262260437,
|
| 727 |
+
"reward_std": 0.2545584440231323,
|
| 728 |
+
"rewards/env_goofspiel_reward/mean": 0.9600000262260437,
|
| 729 |
+
"rewards/env_goofspiel_reward/std": 0.34941941499710083,
|
| 730 |
+
"sampling/importance_sampling_ratio/max": 1.8894808292388916,
|
| 731 |
+
"sampling/importance_sampling_ratio/mean": 0.9526291608810424,
|
| 732 |
+
"sampling/importance_sampling_ratio/min": 0.12061886340379716,
|
| 733 |
+
"sampling/sampling_logp_difference/max": 1.9378384590148925,
|
| 734 |
+
"sampling/sampling_logp_difference/mean": 0.09047931358218193,
|
| 735 |
"step": 105,
|
| 736 |
+
"step_time": 5.383052910200059
|
| 737 |
},
|
| 738 |
{
|
| 739 |
+
"clip_ratio/high_max": 0.002631578966975212,
|
| 740 |
+
"clip_ratio/high_mean": 0.001315789483487606,
|
| 741 |
+
"clip_ratio/low_mean": 0.014272123482078314,
|
| 742 |
+
"clip_ratio/low_min": 0.002631578966975212,
|
| 743 |
+
"clip_ratio/region_mean": 0.01558791296556592,
|
| 744 |
"completions/clipped_ratio": 0.0,
|
| 745 |
+
"completions/max_length": 374.6,
|
| 746 |
+
"completions/max_terminated_length": 374.6,
|
| 747 |
+
"completions/mean_length": 287.1375,
|
| 748 |
+
"completions/mean_terminated_length": 287.1375,
|
| 749 |
+
"completions/min_length": 207.0,
|
| 750 |
+
"completions/min_terminated_length": 207.0,
|
| 751 |
+
"entropy": 0.11877087596803904,
|
| 752 |
"epoch": 0.0176,
|
| 753 |
+
"frac_reward_zero_std": 0.575,
|
| 754 |
+
"grad_norm": 0.03493434935808182,
|
| 755 |
+
"kl": 1.660761559009552,
|
| 756 |
"learning_rate": 9.950352107584324e-06,
|
| 757 |
+
"loss": -0.0012725419364869595,
|
| 758 |
+
"num_tokens": 2950253.0,
|
| 759 |
+
"reward": 0.9823125243186951,
|
| 760 |
+
"reward_std": 0.22300379872322082,
|
| 761 |
+
"rewards/env_goofspiel_reward/mean": 0.9823125243186951,
|
| 762 |
+
"rewards/env_goofspiel_reward/std": 0.34692143797874453,
|
| 763 |
+
"sampling/importance_sampling_ratio/max": 2.0992087841033937,
|
| 764 |
+
"sampling/importance_sampling_ratio/mean": 0.9242552995681763,
|
| 765 |
+
"sampling/importance_sampling_ratio/min": 0.0625721976161003,
|
| 766 |
+
"sampling/sampling_logp_difference/max": 2.7260493278503417,
|
| 767 |
+
"sampling/sampling_logp_difference/mean": 0.11037325486540794,
|
| 768 |
"step": 110,
|
| 769 |
+
"step_time": 5.269659819400658
|
| 770 |
},
|
| 771 |
{
|
| 772 |
+
"clip_ratio/high_max": 0.0029411764815449716,
|
| 773 |
+
"clip_ratio/high_mean": 0.0014705882407724858,
|
| 774 |
+
"clip_ratio/low_mean": 0.010151143837720155,
|
| 775 |
+
"clip_ratio/low_min": 0.003125,
|
| 776 |
+
"clip_ratio/region_mean": 0.011621732078492642,
|
| 777 |
"completions/clipped_ratio": 0.0,
|
| 778 |
+
"completions/max_length": 374.2,
|
| 779 |
+
"completions/max_terminated_length": 374.2,
|
| 780 |
+
"completions/mean_length": 293.24375,
|
| 781 |
+
"completions/mean_terminated_length": 293.24375,
|
| 782 |
+
"completions/min_length": 212.0,
|
| 783 |
+
"completions/min_terminated_length": 212.0,
|
| 784 |
+
"entropy": 0.11200540419667959,
|
| 785 |
"epoch": 0.0184,
|
| 786 |
+
"frac_reward_zero_std": 0.7125,
|
| 787 |
+
"grad_norm": 0.020553065463900566,
|
| 788 |
+
"kl": 1.3526975795626641,
|
| 789 |
"learning_rate": 9.950311889456064e-06,
|
| 790 |
+
"loss": -0.0005503765307366848,
|
| 791 |
+
"num_tokens": 3085187.0,
|
| 792 |
+
"reward": 1.0274375319480895,
|
| 793 |
+
"reward_std": 0.14858081489801406,
|
| 794 |
+
"rewards/env_goofspiel_reward/mean": 1.0274375319480895,
|
| 795 |
+
"rewards/env_goofspiel_reward/std": 0.324732506275177,
|
| 796 |
+
"sampling/importance_sampling_ratio/max": 1.7125876188278197,
|
| 797 |
+
"sampling/importance_sampling_ratio/mean": 0.9561464786529541,
|
| 798 |
+
"sampling/importance_sampling_ratio/min": 0.10175237003713847,
|
| 799 |
+
"sampling/sampling_logp_difference/max": 2.038061809539795,
|
| 800 |
+
"sampling/sampling_logp_difference/mean": 0.07954240888357163,
|
| 801 |
"step": 115,
|
| 802 |
+
"step_time": 5.627786207199643
|
| 803 |
},
|
| 804 |
{
|
| 805 |
"clip_ratio/high_max": 0.0,
|
| 806 |
"clip_ratio/high_mean": 0.0,
|
| 807 |
+
"clip_ratio/low_mean": 0.009591933339834213,
|
| 808 |
+
"clip_ratio/low_min": 0.002500000037252903,
|
| 809 |
+
"clip_ratio/region_mean": 0.009591933339834213,
|
| 810 |
"completions/clipped_ratio": 0.0,
|
| 811 |
+
"completions/max_length": 374.8,
|
| 812 |
+
"completions/max_terminated_length": 374.8,
|
| 813 |
+
"completions/mean_length": 285.71875,
|
| 814 |
+
"completions/mean_terminated_length": 285.71875,
|
| 815 |
+
"completions/min_length": 212.0,
|
| 816 |
+
"completions/min_terminated_length": 212.0,
|
| 817 |
+
"entropy": 0.10979099776595831,
|
| 818 |
"epoch": 0.0192,
|
| 819 |
+
"frac_reward_zero_std": 0.55,
|
| 820 |
+
"grad_norm": 0.0316302515566349,
|
| 821 |
+
"kl": 4.151332414150238,
|
| 822 |
"learning_rate": 9.950269042850943e-06,
|
| 823 |
+
"loss": -0.0010401386767625808,
|
| 824 |
+
"num_tokens": 3218750.0,
|
| 825 |
+
"reward": 1.0012500524520873,
|
| 826 |
+
"reward_std": 0.20682872831821442,
|
| 827 |
+
"rewards/env_goofspiel_reward/mean": 1.0012500524520873,
|
| 828 |
+
"rewards/env_goofspiel_reward/std": 0.3268236696720123,
|
| 829 |
+
"sampling/importance_sampling_ratio/max": 1.743626070022583,
|
| 830 |
+
"sampling/importance_sampling_ratio/mean": 0.9425964832305909,
|
| 831 |
+
"sampling/importance_sampling_ratio/min": 0.10384658798575401,
|
| 832 |
+
"sampling/sampling_logp_difference/max": 2.185199999809265,
|
| 833 |
+
"sampling/sampling_logp_difference/mean": 0.09657576829195022,
|
| 834 |
"step": 120,
|
| 835 |
+
"step_time": 5.24430227939938
|
| 836 |
},
|
| 837 |
{
|
| 838 |
"clip_ratio/high_max": 0.0,
|
| 839 |
"clip_ratio/high_mean": 0.0,
|
| 840 |
+
"clip_ratio/low_mean": 0.005753676500171423,
|
| 841 |
"clip_ratio/low_min": 0.0,
|
| 842 |
+
"clip_ratio/region_mean": 0.005753676500171423,
|
| 843 |
"completions/clipped_ratio": 0.0,
|
| 844 |
+
"completions/max_length": 374.6,
|
| 845 |
+
"completions/max_terminated_length": 374.6,
|
| 846 |
+
"completions/mean_length": 292.33125,
|
| 847 |
+
"completions/mean_terminated_length": 292.33125,
|
| 848 |
+
"completions/min_length": 207.0,
|
| 849 |
+
"completions/min_terminated_length": 207.0,
|
| 850 |
+
"entropy": 0.22348260767757894,
|
| 851 |
"epoch": 0.02,
|
| 852 |
+
"frac_reward_zero_std": 0.4625,
|
| 853 |
+
"grad_norm": 0.046369779855012894,
|
| 854 |
+
"kl": 1.2617603421211243,
|
| 855 |
"learning_rate": 9.95022356779914e-06,
|
| 856 |
+
"loss": -0.0007782015483826399,
|
| 857 |
+
"num_tokens": 3353835.0,
|
| 858 |
+
"reward": 0.8960625171661377,
|
| 859 |
+
"reward_std": 0.2599501311779022,
|
| 860 |
+
"rewards/env_goofspiel_reward/mean": 0.8960625171661377,
|
| 861 |
+
"rewards/env_goofspiel_reward/std": 0.4054258644580841,
|
| 862 |
+
"sampling/importance_sampling_ratio/max": 1.864054226875305,
|
| 863 |
+
"sampling/importance_sampling_ratio/mean": 0.9797484517097473,
|
| 864 |
+
"sampling/importance_sampling_ratio/min": 0.16405612826347352,
|
| 865 |
+
"sampling/sampling_logp_difference/max": 1.281555461883545,
|
| 866 |
+
"sampling/sampling_logp_difference/mean": 0.09155683666467666,
|
| 867 |
"step": 125,
|
| 868 |
+
"step_time": 5.507626709400029
|
| 869 |
},
|
| 870 |
{
|
| 871 |
+
"clip_ratio/high_max": 0.004777777753770351,
|
| 872 |
+
"clip_ratio/high_mean": 0.0032222223468124866,
|
| 873 |
+
"clip_ratio/low_mean": 0.008599499333649873,
|
| 874 |
+
"clip_ratio/low_min": 0.0016129031777381898,
|
| 875 |
+
"clip_ratio/region_mean": 0.01182172168046236,
|
| 876 |
"completions/clipped_ratio": 0.0,
|
| 877 |
+
"completions/max_length": 516.8,
|
| 878 |
+
"completions/max_terminated_length": 516.8,
|
| 879 |
+
"completions/mean_length": 405.09375,
|
| 880 |
+
"completions/mean_terminated_length": 405.09375,
|
| 881 |
+
"completions/min_length": 285.2,
|
| 882 |
+
"completions/min_terminated_length": 285.2,
|
| 883 |
+
"entropy": 0.2538205787539482,
|
| 884 |
"epoch": 0.0208,
|
| 885 |
+
"frac_reward_zero_std": 0.375,
|
| 886 |
+
"grad_norm": 0.03732698783278465,
|
| 887 |
+
"kl": 1.2245262771844865,
|
| 888 |
"learning_rate": 9.950175464332696e-06,
|
| 889 |
+
"loss": -0.001604497991502285,
|
| 890 |
+
"num_tokens": 3506594.0,
|
| 891 |
+
"reward": 0.8659999966621399,
|
| 892 |
+
"reward_std": 0.31071450710296633,
|
| 893 |
+
"rewards/env_goofspiel_reward/mean": 0.8659999966621399,
|
| 894 |
+
"rewards/env_goofspiel_reward/std": 0.4219347298145294,
|
| 895 |
+
"sampling/importance_sampling_ratio/max": 2.1955170392990113,
|
| 896 |
+
"sampling/importance_sampling_ratio/mean": 0.9442144870758057,
|
| 897 |
+
"sampling/importance_sampling_ratio/min": 0.11454989165067672,
|
| 898 |
+
"sampling/sampling_logp_difference/max": 1.5350213170051574,
|
| 899 |
+
"sampling/sampling_logp_difference/mean": 0.10076762735843658,
|
| 900 |
"step": 130,
|
| 901 |
+
"step_time": 6.450475099400319
|
| 902 |
},
|
| 903 |
{
|
| 904 |
"clip_ratio/high_max": 0.0,
|
| 905 |
"clip_ratio/high_mean": 0.0,
|
| 906 |
+
"clip_ratio/low_mean": 0.009226978290826082,
|
| 907 |
+
"clip_ratio/low_min": 0.0018518518656492234,
|
| 908 |
+
"clip_ratio/region_mean": 0.009226978290826082,
|
| 909 |
"completions/clipped_ratio": 0.0,
|
| 910 |
+
"completions/max_length": 732.6,
|
| 911 |
+
"completions/max_terminated_length": 732.6,
|
| 912 |
+
"completions/mean_length": 546.55,
|
| 913 |
+
"completions/mean_terminated_length": 546.55,
|
| 914 |
+
"completions/min_length": 408.0,
|
| 915 |
+
"completions/min_terminated_length": 408.0,
|
| 916 |
+
"entropy": 0.23710975274443627,
|
| 917 |
"epoch": 0.0216,
|
| 918 |
+
"frac_reward_zero_std": 0.3375,
|
| 919 |
+
"grad_norm": 0.06111254543066025,
|
| 920 |
+
"kl": 1.2316449135541916,
|
| 921 |
"learning_rate": 9.950124732485496e-06,
|
| 922 |
+
"loss": -0.0011480608955025673,
|
| 923 |
+
"num_tokens": 3681614.0,
|
| 924 |
+
"reward": 0.8281875014305115,
|
| 925 |
+
"reward_std": 0.2840506821870804,
|
| 926 |
+
"rewards/env_goofspiel_reward/mean": 0.8281875014305115,
|
| 927 |
+
"rewards/env_goofspiel_reward/std": 0.42113742232322693,
|
| 928 |
+
"sampling/importance_sampling_ratio/max": 2.122027373313904,
|
| 929 |
+
"sampling/importance_sampling_ratio/mean": 0.9814165592193603,
|
| 930 |
+
"sampling/importance_sampling_ratio/min": 0.08027622438967227,
|
| 931 |
+
"sampling/sampling_logp_difference/max": 1.9004024505615233,
|
| 932 |
+
"sampling/sampling_logp_difference/mean": 0.08553826659917832,
|
| 933 |
"step": 135,
|
| 934 |
+
"step_time": 7.9832679632005235
|
| 935 |
},
|
| 936 |
{
|
| 937 |
+
"clip_ratio/high_max": 0.003724137879908085,
|
| 938 |
+
"clip_ratio/high_mean": 0.0018620689399540425,
|
| 939 |
+
"clip_ratio/low_mean": 0.004548311699181795,
|
| 940 |
"clip_ratio/low_min": 0.0,
|
| 941 |
+
"clip_ratio/region_mean": 0.006410380732268095,
|
| 942 |
"completions/clipped_ratio": 0.0,
|
| 943 |
+
"completions/max_length": 731.8,
|
| 944 |
+
"completions/max_terminated_length": 731.8,
|
| 945 |
+
"completions/mean_length": 582.2,
|
| 946 |
+
"completions/mean_terminated_length": 582.2,
|
| 947 |
+
"completions/min_length": 408.0,
|
| 948 |
+
"completions/min_terminated_length": 408.0,
|
| 949 |
+
"entropy": 0.1793461933732033,
|
| 950 |
"epoch": 0.0224,
|
| 951 |
+
"frac_reward_zero_std": 0.375,
|
| 952 |
+
"grad_norm": 0.03889832645654678,
|
| 953 |
+
"kl": 2.291236698627472,
|
| 954 |
"learning_rate": 9.95007137229328e-06,
|
| 955 |
+
"loss": -0.0022148976102471353,
|
| 956 |
+
"num_tokens": 3864515.0,
|
| 957 |
+
"reward": 0.9315416216850281,
|
| 958 |
+
"reward_std": 0.3372309923171997,
|
| 959 |
+
"rewards/env_goofspiel_reward/mean": 0.9315416216850281,
|
| 960 |
+
"rewards/env_goofspiel_reward/std": 0.450560998916626,
|
| 961 |
+
"sampling/importance_sampling_ratio/max": 2.064964485168457,
|
| 962 |
+
"sampling/importance_sampling_ratio/mean": 0.9481699466705322,
|
| 963 |
+
"sampling/importance_sampling_ratio/min": 0.06474384777247906,
|
| 964 |
+
"sampling/sampling_logp_difference/max": 2.467247152328491,
|
| 965 |
+
"sampling/sampling_logp_difference/mean": 0.09816154837608337,
|
| 966 |
"step": 140,
|
| 967 |
+
"step_time": 8.106385086600312
|
| 968 |
},
|
| 969 |
{
|
| 970 |
+
"clip_ratio/high_max": 0.005647214874625206,
|
| 971 |
+
"clip_ratio/high_mean": 0.002823607437312603,
|
| 972 |
+
"clip_ratio/low_mean": 0.008411859162151814,
|
| 973 |
+
"clip_ratio/low_min": 0.001923076994717121,
|
| 974 |
+
"clip_ratio/region_mean": 0.011235466599464417,
|
| 975 |
"completions/clipped_ratio": 0.0,
|
| 976 |
+
"completions/max_length": 732.2,
|
| 977 |
+
"completions/max_terminated_length": 732.2,
|
| 978 |
+
"completions/mean_length": 559.74375,
|
| 979 |
+
"completions/mean_terminated_length": 559.74375,
|
| 980 |
+
"completions/min_length": 395.8,
|
| 981 |
+
"completions/min_terminated_length": 395.8,
|
| 982 |
+
"entropy": 0.1993610180914402,
|
| 983 |
"epoch": 0.0232,
|
| 984 |
+
"frac_reward_zero_std": 0.375,
|
| 985 |
+
"grad_norm": 0.031588390469551086,
|
| 986 |
+
"kl": 1.1563616126775742,
|
| 987 |
"learning_rate": 9.950015383793636e-06,
|
| 988 |
+
"loss": -0.002377602644264698,
|
| 989 |
+
"num_tokens": 4041287.0,
|
| 990 |
+
"reward": 0.9530208110809326,
|
| 991 |
+
"reward_std": 0.27621357440948485,
|
| 992 |
+
"rewards/env_goofspiel_reward/mean": 0.9530208110809326,
|
| 993 |
+
"rewards/env_goofspiel_reward/std": 0.40836617946624754,
|
| 994 |
+
"sampling/importance_sampling_ratio/max": 2.1861939907073973,
|
| 995 |
+
"sampling/importance_sampling_ratio/mean": 0.9681573390960694,
|
| 996 |
+
"sampling/importance_sampling_ratio/min": 0.05116618424654007,
|
| 997 |
+
"sampling/sampling_logp_difference/max": 2.082305932044983,
|
| 998 |
+
"sampling/sampling_logp_difference/mean": 0.08064938932657242,
|
| 999 |
"step": 145,
|
| 1000 |
+
"step_time": 7.88830038939941
|
| 1001 |
},
|
| 1002 |
{
|
| 1003 |
+
"clip_ratio/high_max": 0.003750000149011612,
|
| 1004 |
+
"clip_ratio/high_mean": 0.001875000074505806,
|
| 1005 |
+
"clip_ratio/low_mean": 0.00870660375803709,
|
| 1006 |
"clip_ratio/low_min": 0.0,
|
| 1007 |
+
"clip_ratio/region_mean": 0.010581603739410639,
|
| 1008 |
"completions/clipped_ratio": 0.0,
|
| 1009 |
+
"completions/max_length": 730.4,
|
| 1010 |
+
"completions/max_terminated_length": 730.4,
|
| 1011 |
+
"completions/mean_length": 545.94375,
|
| 1012 |
+
"completions/mean_terminated_length": 545.94375,
|
| 1013 |
+
"completions/min_length": 388.4,
|
| 1014 |
+
"completions/min_terminated_length": 388.4,
|
| 1015 |
+
"entropy": 0.17713614292442798,
|
| 1016 |
"epoch": 0.024,
|
| 1017 |
+
"frac_reward_zero_std": 0.3375,
|
| 1018 |
+
"grad_norm": 0.026577942073345184,
|
| 1019 |
+
"kl": 1.0718628287315368,
|
| 1020 |
"learning_rate": 9.949956767026006e-06,
|
| 1021 |
+
"loss": -0.0016981028020381927,
|
| 1022 |
+
"num_tokens": 4215342.0,
|
| 1023 |
+
"reward": 0.9501249670982361,
|
| 1024 |
+
"reward_std": 0.31861052215099334,
|
| 1025 |
+
"rewards/env_goofspiel_reward/mean": 0.9501249670982361,
|
| 1026 |
+
"rewards/env_goofspiel_reward/std": 0.41967723369598386,
|
| 1027 |
+
"sampling/importance_sampling_ratio/max": 2.1912062644958494,
|
| 1028 |
+
"sampling/importance_sampling_ratio/mean": 1.0014893412590027,
|
| 1029 |
+
"sampling/importance_sampling_ratio/min": 0.08548425957560539,
|
| 1030 |
+
"sampling/sampling_logp_difference/max": 1.9367387771606446,
|
| 1031 |
+
"sampling/sampling_logp_difference/mean": 0.07785529419779777,
|
| 1032 |
"step": 150,
|
| 1033 |
+
"step_time": 7.899470061399915
|
| 1034 |
},
|
| 1035 |
{
|
| 1036 |
"epoch": 0.024,
|
|
|
|
| 1040 |
"eval_clip_ratio/low_min": 0.0,
|
| 1041 |
"eval_clip_ratio/region_mean": 0.0,
|
| 1042 |
"eval_completions/clipped_ratio": 0.0,
|
| 1043 |
+
"eval_completions/max_length": 563.4,
|
| 1044 |
+
"eval_completions/max_terminated_length": 563.4,
|
| 1045 |
+
"eval_completions/mean_length": 522.9,
|
| 1046 |
+
"eval_completions/mean_terminated_length": 522.9,
|
| 1047 |
+
"eval_completions/min_length": 483.2,
|
| 1048 |
+
"eval_completions/min_terminated_length": 483.2,
|
| 1049 |
+
"eval_entropy": 0.12081696838140488,
|
| 1050 |
+
"eval_frac_reward_zero_std": 0.7,
|
| 1051 |
+
"eval_kl": 1.9419716477394104,
|
| 1052 |
+
"eval_loss": -0.0004246043972671032,
|
| 1053 |
+
"eval_num_tokens": 4215342.0,
|
| 1054 |
+
"eval_reward": 1.1049999594688416,
|
| 1055 |
+
"eval_reward_std": 0.15320646464824678,
|
| 1056 |
+
"eval_rewards/env_goofspiel_reward/mean": 1.1049999594688416,
|
| 1057 |
+
"eval_rewards/env_goofspiel_reward/std": 0.2007630228996277,
|
| 1058 |
+
"eval_runtime": 3.0894,
|
| 1059 |
+
"eval_samples_per_second": 3.237,
|
| 1060 |
+
"eval_sampling/importance_sampling_ratio/max": 1.3710826635360718,
|
| 1061 |
+
"eval_sampling/importance_sampling_ratio/mean": 0.9819563269615174,
|
| 1062 |
+
"eval_sampling/importance_sampling_ratio/min": 0.5958439499139786,
|
| 1063 |
+
"eval_sampling/sampling_logp_difference/max": 0.8319960534572601,
|
| 1064 |
+
"eval_sampling/sampling_logp_difference/mean": 0.0594001155346632,
|
| 1065 |
+
"eval_steps_per_second": 0.971,
|
| 1066 |
"step": 150
|
| 1067 |
},
|
| 1068 |
{
|
| 1069 |
+
"clip_ratio/high_max": 0.0,
|
| 1070 |
+
"clip_ratio/high_mean": 0.0,
|
| 1071 |
+
"clip_ratio/low_mean": 0.005752738565206528,
|
| 1072 |
+
"clip_ratio/low_min": 0.0,
|
| 1073 |
+
"clip_ratio/region_mean": 0.005752738565206528,
|
| 1074 |
+
"completions/clipped_ratio": 0.0,
|
| 1075 |
+
"completions/max_length": 715.0,
|
| 1076 |
+
"completions/max_terminated_length": 715.0,
|
| 1077 |
+
"completions/mean_length": 553.19375,
|
| 1078 |
+
"completions/mean_terminated_length": 553.19375,
|
| 1079 |
+
"completions/min_length": 422.4,
|
| 1080 |
+
"completions/min_terminated_length": 422.4,
|
| 1081 |
+
"entropy": 0.14132860042154788,
|
| 1082 |
+
"epoch": 0.0248,
|
| 1083 |
+
"frac_reward_zero_std": 0.3375,
|
| 1084 |
+
"grad_norm": 0.09626387804746628,
|
| 1085 |
+
"kl": 2.062944608926773,
|
| 1086 |
+
"learning_rate": 9.949895522031688e-06,
|
| 1087 |
+
"loss": -0.0018732130527496337,
|
| 1088 |
+
"num_tokens": 4391001.0,
|
| 1089 |
+
"reward": 0.9909374356269837,
|
| 1090 |
+
"reward_std": 0.29919453859329226,
|
| 1091 |
+
"rewards/env_goofspiel_reward/mean": 0.9909374356269837,
|
| 1092 |
+
"rewards/env_goofspiel_reward/std": 0.4077799320220947,
|
| 1093 |
+
"sampling/importance_sampling_ratio/max": 2.0831751585006715,
|
| 1094 |
+
"sampling/importance_sampling_ratio/mean": 0.9425314784049987,
|
| 1095 |
+
"sampling/importance_sampling_ratio/min": 0.059898375906050204,
|
| 1096 |
+
"sampling/sampling_logp_difference/max": 2.2892024993896483,
|
| 1097 |
+
"sampling/sampling_logp_difference/mean": 0.08296994566917419,
|
| 1098 |
+
"step": 155,
|
| 1099 |
+
"step_time": 7.558112652399904
|
| 1100 |
+
},
|
| 1101 |
+
{
|
| 1102 |
+
"epoch": 0.02496,
|
| 1103 |
"eval_clip_ratio/high_max": 0.0,
|
| 1104 |
"eval_clip_ratio/high_mean": 0.0,
|
| 1105 |
"eval_clip_ratio/low_mean": 0.0,
|
| 1106 |
"eval_clip_ratio/low_min": 0.0,
|
| 1107 |
"eval_clip_ratio/region_mean": 0.0,
|
| 1108 |
"eval_completions/clipped_ratio": 0.0,
|
| 1109 |
+
"eval_completions/max_length": 563.0,
|
| 1110 |
+
"eval_completions/max_terminated_length": 563.0,
|
| 1111 |
+
"eval_completions/mean_length": 523.1,
|
| 1112 |
+
"eval_completions/mean_terminated_length": 523.1,
|
| 1113 |
+
"eval_completions/min_length": 483.6,
|
| 1114 |
+
"eval_completions/min_terminated_length": 483.6,
|
| 1115 |
+
"eval_entropy": 0.09278524518013001,
|
| 1116 |
+
"eval_frac_reward_zero_std": 0.6,
|
| 1117 |
+
"eval_kl": 1.07927747964859,
|
| 1118 |
+
"eval_loss": 6.382538413163275e-05,
|
| 1119 |
+
"eval_num_tokens": 4427509.0,
|
| 1120 |
+
"eval_reward": 1.1483332872390748,
|
| 1121 |
+
"eval_reward_std": 0.15320646464824678,
|
| 1122 |
+
"eval_rewards/env_goofspiel_reward/mean": 1.1483332872390748,
|
| 1123 |
+
"eval_rewards/env_goofspiel_reward/std": 0.20928734838962554,
|
| 1124 |
+
"eval_runtime": 3.1042,
|
| 1125 |
+
"eval_samples_per_second": 3.221,
|
| 1126 |
+
"eval_sampling/importance_sampling_ratio/max": 1.3094272136688232,
|
| 1127 |
+
"eval_sampling/importance_sampling_ratio/mean": 1.0892619371414185,
|
| 1128 |
+
"eval_sampling/importance_sampling_ratio/min": 0.8267600655555725,
|
| 1129 |
+
"eval_sampling/sampling_logp_difference/max": 0.34806744158267977,
|
| 1130 |
+
"eval_sampling/sampling_logp_difference/mean": 0.03115060944110155,
|
| 1131 |
+
"eval_steps_per_second": 0.966,
|
| 1132 |
+
"step": 156
|
| 1133 |
}
|
| 1134 |
],
|
| 1135 |
"logging_steps": 5,
|
| 1136 |
"max_steps": 18750,
|
| 1137 |
+
"num_input_tokens_seen": 4427509,
|
| 1138 |
"num_train_epochs": 3,
|
| 1139 |
"save_steps": 500,
|
| 1140 |
"stateful_callbacks": {
|
|
|
|
| 1150 |
}
|
| 1151 |
},
|
| 1152 |
"total_flos": 0.0,
|
| 1153 |
+
"train_batch_size": 4,
|
| 1154 |
"trial_name": null,
|
| 1155 |
"trial_params": null
|
| 1156 |
}
|
training_args.bin
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 7185
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:906bc07f18d85f3fdbe47d01e60bbe6f967852d19caecc88d502ce07c5e4aa78
|
| 3 |
size 7185
|