Model save
Browse files- README.md +2 -17
- all_results.json +5 -5
- model-00001-of-00003.safetensors +1 -1
- model-00002-of-00003.safetensors +1 -1
- model-00003-of-00003.safetensors +1 -1
- runs/Jun05_11-04-00_n136-100-194/events.out.tfevents.1717556770.n136-100-194.336424.0 +2 -2
- train_results.json +5 -5
- trainer_state.json +233 -536
README.md
CHANGED
|
@@ -14,16 +14,6 @@ should probably proofread and complete it, then remove this comment. -->
|
|
| 14 |
# zephyr-7b-dpo-full
|
| 15 |
|
| 16 |
This model was trained from scratch on the None dataset.
|
| 17 |
-
It achieves the following results on the evaluation set:
|
| 18 |
-
- Loss: 0.6931
|
| 19 |
-
- Rewards/chosen: -8.4881
|
| 20 |
-
- Rewards/rejected: -8.4881
|
| 21 |
-
- Rewards/accuracies: 0.0
|
| 22 |
-
- Rewards/margins: 0.0
|
| 23 |
-
- Logps/rejected: -164.0651
|
| 24 |
-
- Logps/chosen: -164.0651
|
| 25 |
-
- Logits/rejected: -3.2224
|
| 26 |
-
- Logits/chosen: -3.2224
|
| 27 |
|
| 28 |
## Model description
|
| 29 |
|
|
@@ -48,8 +38,8 @@ The following hyperparameters were used during training:
|
|
| 48 |
- seed: 42
|
| 49 |
- distributed_type: multi-GPU
|
| 50 |
- num_devices: 8
|
| 51 |
-
- gradient_accumulation_steps:
|
| 52 |
-
- total_train_batch_size:
|
| 53 |
- total_eval_batch_size: 64
|
| 54 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 55 |
- lr_scheduler_type: cosine
|
|
@@ -58,11 +48,6 @@ The following hyperparameters were used during training:
|
|
| 58 |
|
| 59 |
### Training results
|
| 60 |
|
| 61 |
-
| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
|
| 62 |
-
|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
|
| 63 |
-
| 0.7723 | 0.29 | 100 | 0.6931 | -2.0247 | -2.0247 | 0.0 | 0.0 | -157.6017 | -157.6017 | -3.1472 | -3.1472 |
|
| 64 |
-
| 0.7717 | 0.57 | 200 | 0.6931 | -12.0830 | -12.0830 | 0.0 | 0.0 | -167.6601 | -167.6601 | -3.1635 | -3.1635 |
|
| 65 |
-
| 0.782 | 0.86 | 300 | 0.6931 | -8.4881 | -8.4881 | 0.0 | 0.0 | -164.0651 | -164.0651 | -3.2224 | -3.2224 |
|
| 66 |
|
| 67 |
|
| 68 |
### Framework versions
|
|
|
|
| 14 |
# zephyr-7b-dpo-full
|
| 15 |
|
| 16 |
This model was trained from scratch on the None dataset.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
## Model description
|
| 19 |
|
|
|
|
| 38 |
- seed: 42
|
| 39 |
- distributed_type: multi-GPU
|
| 40 |
- num_devices: 8
|
| 41 |
+
- gradient_accumulation_steps: 4
|
| 42 |
+
- total_train_batch_size: 256
|
| 43 |
- total_eval_batch_size: 64
|
| 44 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
| 45 |
- lr_scheduler_type: cosine
|
|
|
|
| 48 |
|
| 49 |
### Training results
|
| 50 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 51 |
|
| 52 |
|
| 53 |
### Framework versions
|
all_results.json
CHANGED
|
@@ -1,8 +1,8 @@
|
|
| 1 |
{
|
| 2 |
"epoch": 1.0,
|
| 3 |
-
"train_loss": 0.
|
| 4 |
-
"train_runtime":
|
| 5 |
-
"train_samples":
|
| 6 |
-
"train_samples_per_second":
|
| 7 |
-
"train_steps_per_second": 0.
|
| 8 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"epoch": 1.0,
|
| 3 |
+
"train_loss": 0.42979541909757746,
|
| 4 |
+
"train_runtime": 5368.3646,
|
| 5 |
+
"train_samples": 48530,
|
| 6 |
+
"train_samples_per_second": 9.04,
|
| 7 |
+
"train_steps_per_second": 0.035
|
| 8 |
}
|
model-00001-of-00003.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 4943178720
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:356dc2b1355d662c141aef81f6cc89001c178e4007968f89b8978b8150436157
|
| 3 |
size 4943178720
|
model-00002-of-00003.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 4999819336
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5f1d3e3ba731f6817b54fbce899547aa3234b1ac6c106bb71917516260d9eb90
|
| 3 |
size 4999819336
|
model-00003-of-00003.safetensors
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 4540532728
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:68e0cc8151a7384be178ba6186d9b46fe0d7d5bcec31517e8e6d3b801f63aec5
|
| 3 |
size 4540532728
|
runs/Jun05_11-04-00_n136-100-194/events.out.tfevents.1717556770.n136-100-194.336424.0
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:134fc9b4f27841d895937a26473765b3c2b570f6f97d40ce4d813ecf9ed2916b
|
| 3 |
+
size 18136
|
train_results.json
CHANGED
|
@@ -1,8 +1,8 @@
|
|
| 1 |
{
|
| 2 |
"epoch": 1.0,
|
| 3 |
-
"train_loss": 0.
|
| 4 |
-
"train_runtime":
|
| 5 |
-
"train_samples":
|
| 6 |
-
"train_samples_per_second":
|
| 7 |
-
"train_steps_per_second": 0.
|
| 8 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"epoch": 1.0,
|
| 3 |
+
"train_loss": 0.42979541909757746,
|
| 4 |
+
"train_runtime": 5368.3646,
|
| 5 |
+
"train_samples": 48530,
|
| 6 |
+
"train_samples_per_second": 9.04,
|
| 7 |
+
"train_steps_per_second": 0.035
|
| 8 |
}
|
trainer_state.json
CHANGED
|
@@ -1,22 +1,22 @@
|
|
| 1 |
{
|
| 2 |
"best_metric": null,
|
| 3 |
"best_model_checkpoint": null,
|
| 4 |
-
"epoch":
|
| 5 |
-
"eval_steps":
|
| 6 |
-
"global_step":
|
| 7 |
"is_hyper_param_search": false,
|
| 8 |
"is_local_process_zero": true,
|
| 9 |
"is_world_process_zero": true,
|
| 10 |
"log_history": [
|
| 11 |
{
|
| 12 |
-
"epoch": 0.
|
| 13 |
-
"grad_norm":
|
| 14 |
-
"learning_rate":
|
| 15 |
-
"logits/chosen": -4.
|
| 16 |
-
"logits/rejected": -4.
|
| 17 |
-
"logps/chosen": -
|
| 18 |
-
"logps/rejected": -
|
| 19 |
-
"loss": 0.
|
| 20 |
"rewards/accuracies": 0.0,
|
| 21 |
"rewards/chosen": 0.0,
|
| 22 |
"rewards/margins": 0.0,
|
|
@@ -24,590 +24,287 @@
|
|
| 24 |
"step": 1
|
| 25 |
},
|
| 26 |
{
|
| 27 |
-
"epoch": 0.
|
| 28 |
-
"grad_norm":
|
| 29 |
-
"learning_rate":
|
| 30 |
-
"logits/chosen": -4.
|
| 31 |
-
"logits/rejected": -4.
|
| 32 |
-
"logps/chosen": -
|
| 33 |
-
"logps/rejected": -
|
| 34 |
-
"loss": 0.
|
| 35 |
-
"rewards/accuracies": 0.
|
| 36 |
-
"rewards/chosen": 0.
|
| 37 |
-
"rewards/margins":
|
| 38 |
-
"rewards/rejected": 0.
|
| 39 |
"step": 10
|
| 40 |
},
|
| 41 |
{
|
| 42 |
-
"epoch": 0.
|
| 43 |
-
"grad_norm":
|
| 44 |
-
"learning_rate":
|
| 45 |
-
"logits/chosen": -4.
|
| 46 |
-
"logits/rejected": -4.
|
| 47 |
-
"logps/chosen": -
|
| 48 |
-
"logps/rejected": -
|
| 49 |
-
"loss": 0.
|
| 50 |
-
"rewards/accuracies": 0.
|
| 51 |
-
"rewards/chosen":
|
| 52 |
-
"rewards/margins":
|
| 53 |
-
"rewards/rejected": 0.
|
| 54 |
"step": 20
|
| 55 |
},
|
| 56 |
{
|
| 57 |
-
"epoch": 0.
|
| 58 |
-
"grad_norm":
|
| 59 |
-
"learning_rate": 4.
|
| 60 |
-
"logits/chosen": -4.
|
| 61 |
-
"logits/rejected": -
|
| 62 |
-
"logps/chosen": -
|
| 63 |
-
"logps/rejected": -
|
| 64 |
-
"loss": 0.
|
| 65 |
-
"rewards/accuracies": 0.
|
| 66 |
-
"rewards/chosen":
|
| 67 |
-
"rewards/margins":
|
| 68 |
-
"rewards/rejected":
|
| 69 |
"step": 30
|
| 70 |
},
|
| 71 |
{
|
| 72 |
-
"epoch": 0.
|
| 73 |
-
"grad_norm":
|
| 74 |
-
"learning_rate": 4.
|
| 75 |
-
"logits/chosen": -4.
|
| 76 |
-
"logits/rejected": -
|
| 77 |
-
"logps/chosen": -
|
| 78 |
-
"logps/rejected": -
|
| 79 |
-
"loss": 0.
|
| 80 |
-
"rewards/accuracies": 0.
|
| 81 |
-
"rewards/chosen":
|
| 82 |
-
"rewards/margins":
|
| 83 |
-
"rewards/rejected": -
|
| 84 |
"step": 40
|
| 85 |
},
|
| 86 |
{
|
| 87 |
-
"epoch": 0.
|
| 88 |
-
"grad_norm":
|
| 89 |
-
"learning_rate": 4.
|
| 90 |
-
"logits/chosen": -
|
| 91 |
-
"logits/rejected": -
|
| 92 |
-
"logps/chosen": -
|
| 93 |
-
"logps/rejected": -
|
| 94 |
-
"loss": 0.
|
| 95 |
-
"rewards/accuracies": 0.
|
| 96 |
-
"rewards/chosen":
|
| 97 |
-
"rewards/margins":
|
| 98 |
-
"rewards/rejected": -
|
| 99 |
"step": 50
|
| 100 |
},
|
| 101 |
{
|
| 102 |
-
"epoch": 0.
|
| 103 |
-
"grad_norm":
|
| 104 |
-
"learning_rate": 4.
|
| 105 |
-
"logits/chosen": -
|
| 106 |
-
"logits/rejected": -
|
| 107 |
-
"logps/chosen": -
|
| 108 |
-
"logps/rejected": -
|
| 109 |
-
"loss": 0.
|
| 110 |
-
"rewards/accuracies": 0.
|
| 111 |
-
"rewards/chosen":
|
| 112 |
-
"rewards/margins":
|
| 113 |
-
"rewards/rejected": -
|
| 114 |
"step": 60
|
| 115 |
},
|
| 116 |
{
|
| 117 |
-
"epoch": 0.
|
| 118 |
-
"grad_norm":
|
| 119 |
-
"learning_rate":
|
| 120 |
-
"logits/chosen": -
|
| 121 |
-
"logits/rejected": -
|
| 122 |
-
"logps/chosen": -
|
| 123 |
-
"logps/rejected": -
|
| 124 |
-
"loss": 0.
|
| 125 |
-
"rewards/accuracies": 0.
|
| 126 |
-
"rewards/chosen":
|
| 127 |
-
"rewards/margins":
|
| 128 |
-
"rewards/rejected": -
|
| 129 |
"step": 70
|
| 130 |
},
|
| 131 |
{
|
| 132 |
-
"epoch": 0.
|
| 133 |
-
"grad_norm":
|
| 134 |
-
"learning_rate":
|
| 135 |
-
"logits/chosen": -
|
| 136 |
-
"logits/rejected": -
|
| 137 |
-
"logps/chosen": -
|
| 138 |
-
"logps/rejected": -
|
| 139 |
-
"loss": 0.
|
| 140 |
-
"rewards/accuracies": 0.
|
| 141 |
-
"rewards/chosen":
|
| 142 |
-
"rewards/margins":
|
| 143 |
-
"rewards/rejected": -3.
|
| 144 |
"step": 80
|
| 145 |
},
|
| 146 |
{
|
| 147 |
-
"epoch": 0.
|
| 148 |
-
"grad_norm":
|
| 149 |
-
"learning_rate":
|
| 150 |
-
"logits/chosen": -
|
| 151 |
-
"logits/rejected": -
|
| 152 |
-
"logps/chosen": -
|
| 153 |
-
"logps/rejected": -
|
| 154 |
-
"loss": 0.
|
| 155 |
-
"rewards/accuracies": 0.
|
| 156 |
-
"rewards/chosen":
|
| 157 |
-
"rewards/margins":
|
| 158 |
-
"rewards/rejected": -
|
| 159 |
"step": 90
|
| 160 |
},
|
| 161 |
{
|
| 162 |
-
"epoch": 0.
|
| 163 |
-
"grad_norm":
|
| 164 |
-
"learning_rate":
|
| 165 |
-
"logits/chosen": -
|
| 166 |
-
"logits/rejected": -
|
| 167 |
-
"logps/chosen": -
|
| 168 |
-
"logps/rejected": -
|
| 169 |
-
"loss": 0.
|
| 170 |
-
"rewards/accuracies": 0.
|
| 171 |
-
"rewards/chosen":
|
| 172 |
-
"rewards/margins":
|
| 173 |
-
"rewards/rejected": -
|
| 174 |
-
"step": 100
|
| 175 |
-
},
|
| 176 |
-
{
|
| 177 |
-
"epoch": 0.29,
|
| 178 |
-
"eval_logits/chosen": -3.147157669067383,
|
| 179 |
-
"eval_logits/rejected": -3.147157669067383,
|
| 180 |
-
"eval_logps/chosen": -157.6016845703125,
|
| 181 |
-
"eval_logps/rejected": -157.6016845703125,
|
| 182 |
-
"eval_loss": 0.6931471824645996,
|
| 183 |
-
"eval_rewards/accuracies": 0.0,
|
| 184 |
-
"eval_rewards/chosen": -2.0246658325195312,
|
| 185 |
-
"eval_rewards/margins": 0.0,
|
| 186 |
-
"eval_rewards/rejected": -2.0246658325195312,
|
| 187 |
-
"eval_runtime": 1.5111,
|
| 188 |
-
"eval_samples_per_second": 0.662,
|
| 189 |
-
"eval_steps_per_second": 0.662,
|
| 190 |
"step": 100
|
| 191 |
},
|
| 192 |
{
|
| 193 |
-
"epoch": 0.
|
| 194 |
-
"grad_norm":
|
| 195 |
-
"learning_rate":
|
| 196 |
-
"logits/chosen": -
|
| 197 |
-
"logits/rejected": -
|
| 198 |
-
"logps/chosen": -
|
| 199 |
-
"logps/rejected": -
|
| 200 |
-
"loss": 0.
|
| 201 |
-
"rewards/accuracies": 0.
|
| 202 |
-
"rewards/chosen":
|
| 203 |
-
"rewards/margins":
|
| 204 |
-
"rewards/rejected": -
|
| 205 |
"step": 110
|
| 206 |
},
|
| 207 |
{
|
| 208 |
-
"epoch": 0.
|
| 209 |
-
"grad_norm":
|
| 210 |
-
"learning_rate":
|
| 211 |
-
"logits/chosen": -
|
| 212 |
-
"logits/rejected": -
|
| 213 |
-
"logps/chosen": -
|
| 214 |
-
"logps/rejected": -
|
| 215 |
-
"loss": 0.
|
| 216 |
-
"rewards/accuracies": 0.
|
| 217 |
-
"rewards/chosen":
|
| 218 |
-
"rewards/margins":
|
| 219 |
-
"rewards/rejected": -4.
|
| 220 |
"step": 120
|
| 221 |
},
|
| 222 |
{
|
| 223 |
-
"epoch": 0.
|
| 224 |
-
"grad_norm":
|
| 225 |
-
"learning_rate":
|
| 226 |
-
"logits/chosen": -
|
| 227 |
-
"logits/rejected": -
|
| 228 |
-
"logps/chosen": -
|
| 229 |
-
"logps/rejected": -
|
| 230 |
-
"loss": 0.
|
| 231 |
-
"rewards/accuracies": 0.
|
| 232 |
-
"rewards/chosen":
|
| 233 |
-
"rewards/margins":
|
| 234 |
-
"rewards/rejected": -
|
| 235 |
"step": 130
|
| 236 |
},
|
| 237 |
{
|
| 238 |
-
"epoch": 0.
|
| 239 |
-
"grad_norm":
|
| 240 |
-
"learning_rate":
|
| 241 |
-
"logits/chosen": -
|
| 242 |
-
"logits/rejected": -
|
| 243 |
-
"logps/chosen": -
|
| 244 |
-
"logps/rejected": -
|
| 245 |
-
"loss": 0.
|
| 246 |
-
"rewards/accuracies": 0.
|
| 247 |
-
"rewards/chosen":
|
| 248 |
-
"rewards/margins":
|
| 249 |
-
"rewards/rejected": -
|
| 250 |
"step": 140
|
| 251 |
},
|
| 252 |
{
|
| 253 |
-
"epoch": 0.
|
| 254 |
-
"grad_norm":
|
| 255 |
-
"learning_rate":
|
| 256 |
-
"logits/chosen": -
|
| 257 |
-
"logits/rejected": -
|
| 258 |
-
"logps/chosen": -
|
| 259 |
-
"logps/rejected": -
|
| 260 |
-
"loss": 0.
|
| 261 |
-
"rewards/accuracies": 0.
|
| 262 |
-
"rewards/chosen":
|
| 263 |
-
"rewards/margins":
|
| 264 |
-
"rewards/rejected": -
|
| 265 |
"step": 150
|
| 266 |
},
|
| 267 |
{
|
| 268 |
-
"epoch": 0.
|
| 269 |
-
"grad_norm":
|
| 270 |
-
"learning_rate": 3.
|
| 271 |
-
"logits/chosen": -
|
| 272 |
-
"logits/rejected": -
|
| 273 |
-
"logps/chosen": -
|
| 274 |
-
"logps/rejected": -
|
| 275 |
-
"loss": 0.
|
| 276 |
-
"rewards/accuracies": 0.
|
| 277 |
-
"rewards/chosen":
|
| 278 |
-
"rewards/margins":
|
| 279 |
-
"rewards/rejected": -
|
| 280 |
"step": 160
|
| 281 |
},
|
| 282 |
{
|
| 283 |
-
"epoch": 0.
|
| 284 |
-
"grad_norm":
|
| 285 |
-
"learning_rate":
|
| 286 |
-
"logits/chosen": -
|
| 287 |
-
"logits/rejected": -
|
| 288 |
-
"logps/chosen": -
|
| 289 |
-
"logps/rejected": -
|
| 290 |
-
"loss": 0.
|
| 291 |
-
"rewards/accuracies": 0.
|
| 292 |
-
"rewards/chosen":
|
| 293 |
-
"rewards/margins":
|
| 294 |
-
"rewards/rejected": -
|
| 295 |
"step": 170
|
| 296 |
},
|
| 297 |
{
|
| 298 |
-
"epoch": 0.
|
| 299 |
-
"grad_norm":
|
| 300 |
-
"learning_rate":
|
| 301 |
-
"logits/chosen": -
|
| 302 |
-
"logits/rejected": -
|
| 303 |
-
"logps/chosen": -
|
| 304 |
-
"logps/rejected": -
|
| 305 |
-
"loss": 0.
|
| 306 |
-
"rewards/accuracies": 0.
|
| 307 |
-
"rewards/chosen":
|
| 308 |
-
"rewards/margins":
|
| 309 |
-
"rewards/rejected": -
|
| 310 |
"step": 180
|
| 311 |
},
|
| 312 |
-
{
|
| 313 |
-
"epoch": 0.54,
|
| 314 |
-
"grad_norm": 1756.6999736537189,
|
| 315 |
-
"learning_rate": 2.5623267293451823e-07,
|
| 316 |
-
"logits/chosen": -4.283580303192139,
|
| 317 |
-
"logits/rejected": -4.457066535949707,
|
| 318 |
-
"logps/chosen": -305.109375,
|
| 319 |
-
"logps/rejected": -265.3091125488281,
|
| 320 |
-
"loss": 0.6833,
|
| 321 |
-
"rewards/accuracies": 0.887499988079071,
|
| 322 |
-
"rewards/chosen": 7.185091495513916,
|
| 323 |
-
"rewards/margins": 13.808235168457031,
|
| 324 |
-
"rewards/rejected": -6.623143196105957,
|
| 325 |
-
"step": 190
|
| 326 |
-
},
|
| 327 |
-
{
|
| 328 |
-
"epoch": 0.57,
|
| 329 |
-
"grad_norm": 1748.1359792805438,
|
| 330 |
-
"learning_rate": 2.3131747660339394e-07,
|
| 331 |
-
"logits/chosen": -4.295716285705566,
|
| 332 |
-
"logits/rejected": -4.545838832855225,
|
| 333 |
-
"logps/chosen": -289.03521728515625,
|
| 334 |
-
"logps/rejected": -257.3003234863281,
|
| 335 |
-
"loss": 0.7717,
|
| 336 |
-
"rewards/accuracies": 0.856249988079071,
|
| 337 |
-
"rewards/chosen": 8.096379280090332,
|
| 338 |
-
"rewards/margins": 14.581645011901855,
|
| 339 |
-
"rewards/rejected": -6.485265254974365,
|
| 340 |
-
"step": 200
|
| 341 |
-
},
|
| 342 |
-
{
|
| 343 |
-
"epoch": 0.57,
|
| 344 |
-
"eval_logits/chosen": -3.163522243499756,
|
| 345 |
-
"eval_logits/rejected": -3.163522243499756,
|
| 346 |
-
"eval_logps/chosen": -167.66006469726562,
|
| 347 |
-
"eval_logps/rejected": -167.66006469726562,
|
| 348 |
-
"eval_loss": 0.6931471824645996,
|
| 349 |
-
"eval_rewards/accuracies": 0.0,
|
| 350 |
-
"eval_rewards/chosen": -12.083049774169922,
|
| 351 |
-
"eval_rewards/margins": 0.0,
|
| 352 |
-
"eval_rewards/rejected": -12.083049774169922,
|
| 353 |
-
"eval_runtime": 1.4711,
|
| 354 |
-
"eval_samples_per_second": 0.68,
|
| 355 |
-
"eval_steps_per_second": 0.68,
|
| 356 |
-
"step": 200
|
| 357 |
-
},
|
| 358 |
-
{
|
| 359 |
-
"epoch": 0.6,
|
| 360 |
-
"grad_norm": 1829.602720191424,
|
| 361 |
-
"learning_rate": 2.065879555832674e-07,
|
| 362 |
-
"logits/chosen": -4.250877857208252,
|
| 363 |
-
"logits/rejected": -4.542287826538086,
|
| 364 |
-
"logps/chosen": -303.2696838378906,
|
| 365 |
-
"logps/rejected": -262.7691650390625,
|
| 366 |
-
"loss": 0.8229,
|
| 367 |
-
"rewards/accuracies": 0.8687499761581421,
|
| 368 |
-
"rewards/chosen": 7.95089864730835,
|
| 369 |
-
"rewards/margins": 15.162821769714355,
|
| 370 |
-
"rewards/rejected": -7.211922645568848,
|
| 371 |
-
"step": 210
|
| 372 |
-
},
|
| 373 |
-
{
|
| 374 |
-
"epoch": 0.63,
|
| 375 |
-
"grad_norm": 1788.0044780541311,
|
| 376 |
-
"learning_rate": 1.8228988296424875e-07,
|
| 377 |
-
"logits/chosen": -4.257784366607666,
|
| 378 |
-
"logits/rejected": -4.608870983123779,
|
| 379 |
-
"logps/chosen": -293.54071044921875,
|
| 380 |
-
"logps/rejected": -233.70166015625,
|
| 381 |
-
"loss": 0.7993,
|
| 382 |
-
"rewards/accuracies": 0.8812500238418579,
|
| 383 |
-
"rewards/chosen": 7.973156929016113,
|
| 384 |
-
"rewards/margins": 16.081945419311523,
|
| 385 |
-
"rewards/rejected": -8.108789443969727,
|
| 386 |
-
"step": 220
|
| 387 |
-
},
|
| 388 |
-
{
|
| 389 |
-
"epoch": 0.66,
|
| 390 |
-
"grad_norm": 1900.3878505004946,
|
| 391 |
-
"learning_rate": 1.5866474390840124e-07,
|
| 392 |
-
"logits/chosen": -4.334306716918945,
|
| 393 |
-
"logits/rejected": -4.525221824645996,
|
| 394 |
-
"logps/chosen": -278.41827392578125,
|
| 395 |
-
"logps/rejected": -233.2617645263672,
|
| 396 |
-
"loss": 0.6293,
|
| 397 |
-
"rewards/accuracies": 0.862500011920929,
|
| 398 |
-
"rewards/chosen": 6.539956569671631,
|
| 399 |
-
"rewards/margins": 12.842289924621582,
|
| 400 |
-
"rewards/rejected": -6.30233097076416,
|
| 401 |
-
"step": 230
|
| 402 |
-
},
|
| 403 |
-
{
|
| 404 |
-
"epoch": 0.69,
|
| 405 |
-
"grad_norm": 1759.0935765619065,
|
| 406 |
-
"learning_rate": 1.3594733566170925e-07,
|
| 407 |
-
"logits/chosen": -4.360232353210449,
|
| 408 |
-
"logits/rejected": -4.688153266906738,
|
| 409 |
-
"logps/chosen": -255.55807495117188,
|
| 410 |
-
"logps/rejected": -222.76504516601562,
|
| 411 |
-
"loss": 1.0017,
|
| 412 |
-
"rewards/accuracies": 0.831250011920929,
|
| 413 |
-
"rewards/chosen": 6.956341743469238,
|
| 414 |
-
"rewards/margins": 13.633699417114258,
|
| 415 |
-
"rewards/rejected": -6.677358150482178,
|
| 416 |
-
"step": 240
|
| 417 |
-
},
|
| 418 |
-
{
|
| 419 |
-
"epoch": 0.71,
|
| 420 |
-
"grad_norm": 1730.0292892661773,
|
| 421 |
-
"learning_rate": 1.1436343403356016e-07,
|
| 422 |
-
"logits/chosen": -4.352273941040039,
|
| 423 |
-
"logits/rejected": -4.6365180015563965,
|
| 424 |
-
"logps/chosen": -276.07659912109375,
|
| 425 |
-
"logps/rejected": -241.01425170898438,
|
| 426 |
-
"loss": 0.8087,
|
| 427 |
-
"rewards/accuracies": 0.8687499761581421,
|
| 428 |
-
"rewards/chosen": 6.498913764953613,
|
| 429 |
-
"rewards/margins": 14.106257438659668,
|
| 430 |
-
"rewards/rejected": -7.6073455810546875,
|
| 431 |
-
"step": 250
|
| 432 |
-
},
|
| 433 |
-
{
|
| 434 |
-
"epoch": 0.74,
|
| 435 |
-
"grad_norm": 1731.9100006560693,
|
| 436 |
-
"learning_rate": 9.412754953531663e-08,
|
| 437 |
-
"logits/chosen": -4.388053894042969,
|
| 438 |
-
"logits/rejected": -4.64432954788208,
|
| 439 |
-
"logps/chosen": -270.12225341796875,
|
| 440 |
-
"logps/rejected": -249.11416625976562,
|
| 441 |
-
"loss": 0.6771,
|
| 442 |
-
"rewards/accuracies": 0.8687499761581421,
|
| 443 |
-
"rewards/chosen": 5.336110591888428,
|
| 444 |
-
"rewards/margins": 13.311798095703125,
|
| 445 |
-
"rewards/rejected": -7.975686073303223,
|
| 446 |
-
"step": 260
|
| 447 |
-
},
|
| 448 |
-
{
|
| 449 |
-
"epoch": 0.77,
|
| 450 |
-
"grad_norm": 1559.894108247427,
|
| 451 |
-
"learning_rate": 7.544079547848181e-08,
|
| 452 |
-
"logits/chosen": -4.511970520019531,
|
| 453 |
-
"logits/rejected": -4.677350044250488,
|
| 454 |
-
"logps/chosen": -272.5389709472656,
|
| 455 |
-
"logps/rejected": -237.0705108642578,
|
| 456 |
-
"loss": 0.816,
|
| 457 |
-
"rewards/accuracies": 0.831250011920929,
|
| 458 |
-
"rewards/chosen": 4.5388288497924805,
|
| 459 |
-
"rewards/margins": 12.382614135742188,
|
| 460 |
-
"rewards/rejected": -7.843785762786865,
|
| 461 |
-
"step": 270
|
| 462 |
-
},
|
| 463 |
-
{
|
| 464 |
-
"epoch": 0.8,
|
| 465 |
-
"grad_norm": 2131.975353118901,
|
| 466 |
-
"learning_rate": 5.848888922025552e-08,
|
| 467 |
-
"logits/chosen": -4.293630123138428,
|
| 468 |
-
"logits/rejected": -4.587409496307373,
|
| 469 |
-
"logps/chosen": -272.18914794921875,
|
| 470 |
-
"logps/rejected": -237.04507446289062,
|
| 471 |
-
"loss": 0.7076,
|
| 472 |
-
"rewards/accuracies": 0.8500000238418579,
|
| 473 |
-
"rewards/chosen": 6.89593505859375,
|
| 474 |
-
"rewards/margins": 13.17906379699707,
|
| 475 |
-
"rewards/rejected": -6.283128261566162,
|
| 476 |
-
"step": 280
|
| 477 |
-
},
|
| 478 |
-
{
|
| 479 |
-
"epoch": 0.83,
|
| 480 |
-
"grad_norm": 2073.5868352302195,
|
| 481 |
-
"learning_rate": 4.3440306421001324e-08,
|
| 482 |
-
"logits/chosen": -4.306157112121582,
|
| 483 |
-
"logits/rejected": -4.501837253570557,
|
| 484 |
-
"logps/chosen": -279.41571044921875,
|
| 485 |
-
"logps/rejected": -245.3014678955078,
|
| 486 |
-
"loss": 0.7257,
|
| 487 |
-
"rewards/accuracies": 0.918749988079071,
|
| 488 |
-
"rewards/chosen": 7.150078773498535,
|
| 489 |
-
"rewards/margins": 15.104804992675781,
|
| 490 |
-
"rewards/rejected": -7.954724311828613,
|
| 491 |
-
"step": 290
|
| 492 |
-
},
|
| 493 |
-
{
|
| 494 |
-
"epoch": 0.86,
|
| 495 |
-
"grad_norm": 1664.7949029760775,
|
| 496 |
-
"learning_rate": 3.044460665744283e-08,
|
| 497 |
-
"logits/chosen": -4.291565418243408,
|
| 498 |
-
"logits/rejected": -4.546942234039307,
|
| 499 |
-
"logps/chosen": -294.171142578125,
|
| 500 |
-
"logps/rejected": -248.09219360351562,
|
| 501 |
-
"loss": 0.782,
|
| 502 |
-
"rewards/accuracies": 0.875,
|
| 503 |
-
"rewards/chosen": 8.660406112670898,
|
| 504 |
-
"rewards/margins": 15.387075424194336,
|
| 505 |
-
"rewards/rejected": -6.726672172546387,
|
| 506 |
-
"step": 300
|
| 507 |
-
},
|
| 508 |
-
{
|
| 509 |
-
"epoch": 0.86,
|
| 510 |
-
"eval_logits/chosen": -3.222372531890869,
|
| 511 |
-
"eval_logits/rejected": -3.222372531890869,
|
| 512 |
-
"eval_logps/chosen": -164.06509399414062,
|
| 513 |
-
"eval_logps/rejected": -164.06509399414062,
|
| 514 |
-
"eval_loss": 0.6931471824645996,
|
| 515 |
-
"eval_rewards/accuracies": 0.0,
|
| 516 |
-
"eval_rewards/chosen": -8.488082885742188,
|
| 517 |
-
"eval_rewards/margins": 0.0,
|
| 518 |
-
"eval_rewards/rejected": -8.488082885742188,
|
| 519 |
-
"eval_runtime": 1.4741,
|
| 520 |
-
"eval_samples_per_second": 0.678,
|
| 521 |
-
"eval_steps_per_second": 0.678,
|
| 522 |
-
"step": 300
|
| 523 |
-
},
|
| 524 |
-
{
|
| 525 |
-
"epoch": 0.89,
|
| 526 |
-
"grad_norm": 2673.859330983399,
|
| 527 |
-
"learning_rate": 1.9630947032398066e-08,
|
| 528 |
-
"logits/chosen": -4.44521427154541,
|
| 529 |
-
"logits/rejected": -4.5893964767456055,
|
| 530 |
-
"logps/chosen": -260.0516662597656,
|
| 531 |
-
"logps/rejected": -234.23690795898438,
|
| 532 |
-
"loss": 0.7257,
|
| 533 |
-
"rewards/accuracies": 0.856249988079071,
|
| 534 |
-
"rewards/chosen": 6.064515113830566,
|
| 535 |
-
"rewards/margins": 13.008413314819336,
|
| 536 |
-
"rewards/rejected": -6.943899631500244,
|
| 537 |
-
"step": 310
|
| 538 |
-
},
|
| 539 |
-
{
|
| 540 |
-
"epoch": 0.91,
|
| 541 |
-
"grad_norm": 1691.3211570500287,
|
| 542 |
-
"learning_rate": 1.1106798553464802e-08,
|
| 543 |
-
"logits/chosen": -4.273613929748535,
|
| 544 |
-
"logits/rejected": -4.540968894958496,
|
| 545 |
-
"logps/chosen": -289.5185546875,
|
| 546 |
-
"logps/rejected": -251.7852325439453,
|
| 547 |
-
"loss": 0.6936,
|
| 548 |
-
"rewards/accuracies": 0.887499988079071,
|
| 549 |
-
"rewards/chosen": 7.365248203277588,
|
| 550 |
-
"rewards/margins": 14.206552505493164,
|
| 551 |
-
"rewards/rejected": -6.84130334854126,
|
| 552 |
-
"step": 320
|
| 553 |
-
},
|
| 554 |
-
{
|
| 555 |
-
"epoch": 0.94,
|
| 556 |
-
"grad_norm": 2384.133770310185,
|
| 557 |
-
"learning_rate": 4.956878037864043e-09,
|
| 558 |
-
"logits/chosen": -4.335446834564209,
|
| 559 |
-
"logits/rejected": -4.543330669403076,
|
| 560 |
-
"logps/chosen": -303.6029968261719,
|
| 561 |
-
"logps/rejected": -260.98760986328125,
|
| 562 |
-
"loss": 0.7573,
|
| 563 |
-
"rewards/accuracies": 0.856249988079071,
|
| 564 |
-
"rewards/chosen": 5.982678413391113,
|
| 565 |
-
"rewards/margins": 13.970565795898438,
|
| 566 |
-
"rewards/rejected": -7.987887382507324,
|
| 567 |
-
"step": 330
|
| 568 |
-
},
|
| 569 |
-
{
|
| 570 |
-
"epoch": 0.97,
|
| 571 |
-
"grad_norm": 2211.192973189294,
|
| 572 |
-
"learning_rate": 1.2423061586496476e-09,
|
| 573 |
-
"logits/chosen": -4.319240093231201,
|
| 574 |
-
"logits/rejected": -4.623973846435547,
|
| 575 |
-
"logps/chosen": -283.38006591796875,
|
| 576 |
-
"logps/rejected": -234.4169921875,
|
| 577 |
-
"loss": 0.8286,
|
| 578 |
-
"rewards/accuracies": 0.8687499761581421,
|
| 579 |
-
"rewards/chosen": 5.565188407897949,
|
| 580 |
-
"rewards/margins": 11.47750186920166,
|
| 581 |
-
"rewards/rejected": -5.912313938140869,
|
| 582 |
-
"step": 340
|
| 583 |
-
},
|
| 584 |
-
{
|
| 585 |
-
"epoch": 1.0,
|
| 586 |
-
"grad_norm": 1586.2152634156816,
|
| 587 |
-
"learning_rate": 0.0,
|
| 588 |
-
"logits/chosen": -4.285008430480957,
|
| 589 |
-
"logits/rejected": -4.503040790557861,
|
| 590 |
-
"logps/chosen": -289.0204162597656,
|
| 591 |
-
"logps/rejected": -251.9452362060547,
|
| 592 |
-
"loss": 0.7666,
|
| 593 |
-
"rewards/accuracies": 0.875,
|
| 594 |
-
"rewards/chosen": 7.000736236572266,
|
| 595 |
-
"rewards/margins": 13.714078903198242,
|
| 596 |
-
"rewards/rejected": -6.713343143463135,
|
| 597 |
-
"step": 350
|
| 598 |
-
},
|
| 599 |
{
|
| 600 |
"epoch": 1.0,
|
| 601 |
-
"step":
|
| 602 |
"total_flos": 0.0,
|
| 603 |
-
"train_loss": 0.
|
| 604 |
-
"train_runtime":
|
| 605 |
-
"train_samples_per_second":
|
| 606 |
-
"train_steps_per_second": 0.
|
| 607 |
}
|
| 608 |
],
|
| 609 |
"logging_steps": 10,
|
| 610 |
-
"max_steps":
|
| 611 |
"num_input_tokens_seen": 0,
|
| 612 |
"num_train_epochs": 1,
|
| 613 |
"save_steps": 100,
|
|
|
|
| 1 |
{
|
| 2 |
"best_metric": null,
|
| 3 |
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 0.9960474308300395,
|
| 5 |
+
"eval_steps": 500,
|
| 6 |
+
"global_step": 189,
|
| 7 |
"is_hyper_param_search": false,
|
| 8 |
"is_local_process_zero": true,
|
| 9 |
"is_world_process_zero": true,
|
| 10 |
"log_history": [
|
| 11 |
{
|
| 12 |
+
"epoch": 0.01,
|
| 13 |
+
"grad_norm": 26.908694644642612,
|
| 14 |
+
"learning_rate": 2.6315789473684208e-08,
|
| 15 |
+
"logits/chosen": -4.638427734375,
|
| 16 |
+
"logits/rejected": -4.891327857971191,
|
| 17 |
+
"logps/chosen": -198.52749633789062,
|
| 18 |
+
"logps/rejected": -147.3392791748047,
|
| 19 |
+
"loss": 0.6929,
|
| 20 |
"rewards/accuracies": 0.0,
|
| 21 |
"rewards/chosen": 0.0,
|
| 22 |
"rewards/margins": 0.0,
|
|
|
|
| 24 |
"step": 1
|
| 25 |
},
|
| 26 |
{
|
| 27 |
+
"epoch": 0.05,
|
| 28 |
+
"grad_norm": 27.07664091604203,
|
| 29 |
+
"learning_rate": 2.631578947368421e-07,
|
| 30 |
+
"logits/chosen": -4.496448993682861,
|
| 31 |
+
"logits/rejected": -4.815927028656006,
|
| 32 |
+
"logps/chosen": -224.28125,
|
| 33 |
+
"logps/rejected": -167.94735717773438,
|
| 34 |
+
"loss": 0.6915,
|
| 35 |
+
"rewards/accuracies": 0.5034722089767456,
|
| 36 |
+
"rewards/chosen": 0.0034646072890609503,
|
| 37 |
+
"rewards/margins": 0.002639756305143237,
|
| 38 |
+
"rewards/rejected": 0.0008248506928794086,
|
| 39 |
"step": 10
|
| 40 |
},
|
| 41 |
{
|
| 42 |
+
"epoch": 0.11,
|
| 43 |
+
"grad_norm": 26.30402064096193,
|
| 44 |
+
"learning_rate": 4.999573126145131e-07,
|
| 45 |
+
"logits/chosen": -4.625959873199463,
|
| 46 |
+
"logits/rejected": -4.94482421875,
|
| 47 |
+
"logps/chosen": -231.04525756835938,
|
| 48 |
+
"logps/rejected": -196.3661651611328,
|
| 49 |
+
"loss": 0.6465,
|
| 50 |
+
"rewards/accuracies": 0.8125,
|
| 51 |
+
"rewards/chosen": -0.04837086424231529,
|
| 52 |
+
"rewards/margins": 0.11359457671642303,
|
| 53 |
+
"rewards/rejected": -0.16196544468402863,
|
| 54 |
"step": 20
|
| 55 |
},
|
| 56 |
{
|
| 57 |
+
"epoch": 0.16,
|
| 58 |
+
"grad_norm": 25.70165553073792,
|
| 59 |
+
"learning_rate": 4.948524419003415e-07,
|
| 60 |
+
"logits/chosen": -4.869608402252197,
|
| 61 |
+
"logits/rejected": -5.148451805114746,
|
| 62 |
+
"logps/chosen": -273.7060241699219,
|
| 63 |
+
"logps/rejected": -259.2108154296875,
|
| 64 |
+
"loss": 0.5717,
|
| 65 |
+
"rewards/accuracies": 0.778124988079071,
|
| 66 |
+
"rewards/chosen": -0.5196550488471985,
|
| 67 |
+
"rewards/margins": 0.3445150554180145,
|
| 68 |
+
"rewards/rejected": -0.8641700744628906,
|
| 69 |
"step": 30
|
| 70 |
},
|
| 71 |
{
|
| 72 |
+
"epoch": 0.21,
|
| 73 |
+
"grad_norm": 32.38040367732233,
|
| 74 |
+
"learning_rate": 4.81409414945389e-07,
|
| 75 |
+
"logits/chosen": -4.95624303817749,
|
| 76 |
+
"logits/rejected": -5.334275245666504,
|
| 77 |
+
"logps/chosen": -321.26739501953125,
|
| 78 |
+
"logps/rejected": -317.9222106933594,
|
| 79 |
+
"loss": 0.5311,
|
| 80 |
+
"rewards/accuracies": 0.793749988079071,
|
| 81 |
+
"rewards/chosen": -0.9283856153488159,
|
| 82 |
+
"rewards/margins": 0.48562851548194885,
|
| 83 |
+
"rewards/rejected": -1.4140141010284424,
|
| 84 |
"step": 40
|
| 85 |
},
|
| 86 |
{
|
| 87 |
+
"epoch": 0.26,
|
| 88 |
+
"grad_norm": 27.107807886309228,
|
| 89 |
+
"learning_rate": 4.6008601790947314e-07,
|
| 90 |
+
"logits/chosen": -5.323241233825684,
|
| 91 |
+
"logits/rejected": -5.817015171051025,
|
| 92 |
+
"logps/chosen": -357.8787536621094,
|
| 93 |
+
"logps/rejected": -385.47576904296875,
|
| 94 |
+
"loss": 0.4831,
|
| 95 |
+
"rewards/accuracies": 0.8500000238418579,
|
| 96 |
+
"rewards/chosen": -1.411299467086792,
|
| 97 |
+
"rewards/margins": 0.7530988454818726,
|
| 98 |
+
"rewards/rejected": -2.164398193359375,
|
| 99 |
"step": 50
|
| 100 |
},
|
| 101 |
{
|
| 102 |
+
"epoch": 0.32,
|
| 103 |
+
"grad_norm": 32.232061879934236,
|
| 104 |
+
"learning_rate": 4.3160839350405605e-07,
|
| 105 |
+
"logits/chosen": -5.831389904022217,
|
| 106 |
+
"logits/rejected": -6.2499542236328125,
|
| 107 |
+
"logps/chosen": -395.7707824707031,
|
| 108 |
+
"logps/rejected": -446.3265686035156,
|
| 109 |
+
"loss": 0.4294,
|
| 110 |
+
"rewards/accuracies": 0.778124988079071,
|
| 111 |
+
"rewards/chosen": -1.821434736251831,
|
| 112 |
+
"rewards/margins": 0.9432821273803711,
|
| 113 |
+
"rewards/rejected": -2.7647171020507812,
|
| 114 |
"step": 60
|
| 115 |
},
|
| 116 |
{
|
| 117 |
+
"epoch": 0.37,
|
| 118 |
+
"grad_norm": 35.03072007251475,
|
| 119 |
+
"learning_rate": 3.9694631307311825e-07,
|
| 120 |
+
"logits/chosen": -6.090306758880615,
|
| 121 |
+
"logits/rejected": -6.541258335113525,
|
| 122 |
+
"logps/chosen": -430.2369689941406,
|
| 123 |
+
"logps/rejected": -496.2119140625,
|
| 124 |
+
"loss": 0.424,
|
| 125 |
+
"rewards/accuracies": 0.8187500238418579,
|
| 126 |
+
"rewards/chosen": -2.1488282680511475,
|
| 127 |
+
"rewards/margins": 1.0692826509475708,
|
| 128 |
+
"rewards/rejected": -3.218111038208008,
|
| 129 |
"step": 70
|
| 130 |
},
|
| 131 |
{
|
| 132 |
+
"epoch": 0.42,
|
| 133 |
+
"grad_norm": 30.667469826354093,
|
| 134 |
+
"learning_rate": 3.572801521931522e-07,
|
| 135 |
+
"logits/chosen": -6.3887619972229,
|
| 136 |
+
"logits/rejected": -6.877404689788818,
|
| 137 |
+
"logps/chosen": -439.2911071777344,
|
| 138 |
+
"logps/rejected": -526.5487060546875,
|
| 139 |
+
"loss": 0.4001,
|
| 140 |
+
"rewards/accuracies": 0.815625011920929,
|
| 141 |
+
"rewards/chosen": -2.3115358352661133,
|
| 142 |
+
"rewards/margins": 1.2141239643096924,
|
| 143 |
+
"rewards/rejected": -3.5256600379943848,
|
| 144 |
"step": 80
|
| 145 |
},
|
| 146 |
{
|
| 147 |
+
"epoch": 0.47,
|
| 148 |
+
"grad_norm": 37.36819911889553,
|
| 149 |
+
"learning_rate": 3.139606943986089e-07,
|
| 150 |
+
"logits/chosen": -6.5696258544921875,
|
| 151 |
+
"logits/rejected": -7.1035637855529785,
|
| 152 |
+
"logps/chosen": -458.3387756347656,
|
| 153 |
+
"logps/rejected": -556.1650390625,
|
| 154 |
+
"loss": 0.3875,
|
| 155 |
+
"rewards/accuracies": 0.7718750238418579,
|
| 156 |
+
"rewards/chosen": -2.5067451000213623,
|
| 157 |
+
"rewards/margins": 1.3494058847427368,
|
| 158 |
+
"rewards/rejected": -3.8561508655548096,
|
| 159 |
"step": 90
|
| 160 |
},
|
| 161 |
{
|
| 162 |
+
"epoch": 0.53,
|
| 163 |
+
"grad_norm": 33.15053822353323,
|
| 164 |
+
"learning_rate": 2.684631318687185e-07,
|
| 165 |
+
"logits/chosen": -6.621747016906738,
|
| 166 |
+
"logits/rejected": -7.236710548400879,
|
| 167 |
+
"logps/chosen": -467.0467834472656,
|
| 168 |
+
"logps/rejected": -582.046142578125,
|
| 169 |
+
"loss": 0.3867,
|
| 170 |
+
"rewards/accuracies": 0.796875,
|
| 171 |
+
"rewards/chosen": -2.4837827682495117,
|
| 172 |
+
"rewards/margins": 1.5418504476547241,
|
| 173 |
+
"rewards/rejected": -4.025633811950684,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 174 |
"step": 100
|
| 175 |
},
|
| 176 |
{
|
| 177 |
+
"epoch": 0.58,
|
| 178 |
+
"grad_norm": 37.45830028947681,
|
| 179 |
+
"learning_rate": 2.2233682952712483e-07,
|
| 180 |
+
"logits/chosen": -6.568659782409668,
|
| 181 |
+
"logits/rejected": -7.284300327301025,
|
| 182 |
+
"logps/chosen": -460.4766540527344,
|
| 183 |
+
"logps/rejected": -578.6600341796875,
|
| 184 |
+
"loss": 0.3771,
|
| 185 |
+
"rewards/accuracies": 0.846875011920929,
|
| 186 |
+
"rewards/chosen": -2.3609726428985596,
|
| 187 |
+
"rewards/margins": 1.647943139076233,
|
| 188 |
+
"rewards/rejected": -4.008915901184082,
|
| 189 |
"step": 110
|
| 190 |
},
|
| 191 |
{
|
| 192 |
+
"epoch": 0.63,
|
| 193 |
+
"grad_norm": 33.8427535333109,
|
| 194 |
+
"learning_rate": 1.7715256327766884e-07,
|
| 195 |
+
"logits/chosen": -6.796021461486816,
|
| 196 |
+
"logits/rejected": -7.497170925140381,
|
| 197 |
+
"logps/chosen": -504.50543212890625,
|
| 198 |
+
"logps/rejected": -621.22314453125,
|
| 199 |
+
"loss": 0.3508,
|
| 200 |
+
"rewards/accuracies": 0.815625011920929,
|
| 201 |
+
"rewards/chosen": -2.883434295654297,
|
| 202 |
+
"rewards/margins": 1.6248239278793335,
|
| 203 |
+
"rewards/rejected": -4.50825834274292,
|
| 204 |
"step": 120
|
| 205 |
},
|
| 206 |
{
|
| 207 |
+
"epoch": 0.69,
|
| 208 |
+
"grad_norm": 35.353347844932394,
|
| 209 |
+
"learning_rate": 1.3444902911492174e-07,
|
| 210 |
+
"logits/chosen": -6.833544731140137,
|
| 211 |
+
"logits/rejected": -7.472651481628418,
|
| 212 |
+
"logps/chosen": -521.9656372070312,
|
| 213 |
+
"logps/rejected": -659.3110961914062,
|
| 214 |
+
"loss": 0.3705,
|
| 215 |
+
"rewards/accuracies": 0.840624988079071,
|
| 216 |
+
"rewards/chosen": -2.989759922027588,
|
| 217 |
+
"rewards/margins": 1.8119176626205444,
|
| 218 |
+
"rewards/rejected": -4.801677227020264,
|
| 219 |
"step": 130
|
| 220 |
},
|
| 221 |
{
|
| 222 |
+
"epoch": 0.74,
|
| 223 |
+
"grad_norm": 32.6045025544378,
|
| 224 |
+
"learning_rate": 9.56804446775518e-08,
|
| 225 |
+
"logits/chosen": -6.738868713378906,
|
| 226 |
+
"logits/rejected": -7.498864650726318,
|
| 227 |
+
"logps/chosen": -470.77337646484375,
|
| 228 |
+
"logps/rejected": -584.4710083007812,
|
| 229 |
+
"loss": 0.3591,
|
| 230 |
+
"rewards/accuracies": 0.831250011920929,
|
| 231 |
+
"rewards/chosen": -2.558176040649414,
|
| 232 |
+
"rewards/margins": 1.6057535409927368,
|
| 233 |
+
"rewards/rejected": -4.163929462432861,
|
| 234 |
"step": 140
|
| 235 |
},
|
| 236 |
{
|
| 237 |
+
"epoch": 0.79,
|
| 238 |
+
"grad_norm": 32.49183208247093,
|
| 239 |
+
"learning_rate": 6.216702761078166e-08,
|
| 240 |
+
"logits/chosen": -7.049106597900391,
|
| 241 |
+
"logits/rejected": -7.772597312927246,
|
| 242 |
+
"logps/chosen": -487.25726318359375,
|
| 243 |
+
"logps/rejected": -619.6534423828125,
|
| 244 |
+
"loss": 0.3576,
|
| 245 |
+
"rewards/accuracies": 0.8343750238418579,
|
| 246 |
+
"rewards/chosen": -2.82848858833313,
|
| 247 |
+
"rewards/margins": 1.7726719379425049,
|
| 248 |
+
"rewards/rejected": -4.601161003112793,
|
| 249 |
"step": 150
|
| 250 |
},
|
| 251 |
{
|
| 252 |
+
"epoch": 0.84,
|
| 253 |
+
"grad_norm": 37.26747220029015,
|
| 254 |
+
"learning_rate": 3.5050037137906885e-08,
|
| 255 |
+
"logits/chosen": -6.9701337814331055,
|
| 256 |
+
"logits/rejected": -7.731366157531738,
|
| 257 |
+
"logps/chosen": -494.5716247558594,
|
| 258 |
+
"logps/rejected": -623.4630737304688,
|
| 259 |
+
"loss": 0.3502,
|
| 260 |
+
"rewards/accuracies": 0.84375,
|
| 261 |
+
"rewards/chosen": -2.750206470489502,
|
| 262 |
+
"rewards/margins": 1.7980148792266846,
|
| 263 |
+
"rewards/rejected": -4.548220634460449,
|
| 264 |
"step": 160
|
| 265 |
},
|
| 266 |
{
|
| 267 |
+
"epoch": 0.9,
|
| 268 |
+
"grad_norm": 31.918546112926368,
|
| 269 |
+
"learning_rate": 1.5252909846235894e-08,
|
| 270 |
+
"logits/chosen": -7.007571220397949,
|
| 271 |
+
"logits/rejected": -7.6982011795043945,
|
| 272 |
+
"logps/chosen": -509.54388427734375,
|
| 273 |
+
"logps/rejected": -666.7489624023438,
|
| 274 |
+
"loss": 0.3631,
|
| 275 |
+
"rewards/accuracies": 0.890625,
|
| 276 |
+
"rewards/chosen": -2.9002063274383545,
|
| 277 |
+
"rewards/margins": 2.0059866905212402,
|
| 278 |
+
"rewards/rejected": -4.906193733215332,
|
| 279 |
"step": 170
|
| 280 |
},
|
| 281 |
{
|
| 282 |
+
"epoch": 0.95,
|
| 283 |
+
"grad_norm": 29.32551345390984,
|
| 284 |
+
"learning_rate": 3.4498131616493565e-09,
|
| 285 |
+
"logits/chosen": -6.939836025238037,
|
| 286 |
+
"logits/rejected": -7.576680660247803,
|
| 287 |
+
"logps/chosen": -514.7128295898438,
|
| 288 |
+
"logps/rejected": -656.9924926757812,
|
| 289 |
+
"loss": 0.3518,
|
| 290 |
+
"rewards/accuracies": 0.8031250238418579,
|
| 291 |
+
"rewards/chosen": -2.932391881942749,
|
| 292 |
+
"rewards/margins": 1.7769733667373657,
|
| 293 |
+
"rewards/rejected": -4.709364891052246,
|
| 294 |
"step": 180
|
| 295 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 296 |
{
|
| 297 |
"epoch": 1.0,
|
| 298 |
+
"step": 189,
|
| 299 |
"total_flos": 0.0,
|
| 300 |
+
"train_loss": 0.42979541909757746,
|
| 301 |
+
"train_runtime": 5368.3646,
|
| 302 |
+
"train_samples_per_second": 9.04,
|
| 303 |
+
"train_steps_per_second": 0.035
|
| 304 |
}
|
| 305 |
],
|
| 306 |
"logging_steps": 10,
|
| 307 |
+
"max_steps": 189,
|
| 308 |
"num_input_tokens_seen": 0,
|
| 309 |
"num_train_epochs": 1,
|
| 310 |
"save_steps": 100,
|