Model save
Browse files- README.md +12 -12
- all_results.json +4 -4
- config.json +2 -2
- model-00001-of-00003.safetensors +1 -1
- model-00002-of-00003.safetensors +1 -1
- model-00003-of-00003.safetensors +1 -1
- runs/May29_00-49-01_n136-129-074/events.out.tfevents.1716915281.n136-129-074.1758650.0 +2 -2
- tokenizer.json +0 -1
- train_results.json +4 -4
- trainer_state.json +1040 -500
- training_args.bin +2 -2
README.md
CHANGED
|
@@ -15,15 +15,15 @@ should probably proofread and complete it, then remove this comment. -->
|
|
| 15 |
|
| 16 |
This model was trained from scratch on the None dataset.
|
| 17 |
It achieves the following results on the evaluation set:
|
| 18 |
-
- Loss: 1.
|
| 19 |
-
- Rewards/chosen: -
|
| 20 |
-
- Rewards/rejected: -5.
|
| 21 |
-
- Rewards/accuracies: 0.
|
| 22 |
-
- Rewards/margins: 0.
|
| 23 |
-
- Logps/rejected: -
|
| 24 |
-
- Logps/chosen: -
|
| 25 |
-
- Logits/rejected: -4.
|
| 26 |
-
- Logits/chosen: -4.
|
| 27 |
|
| 28 |
## Model description
|
| 29 |
|
|
@@ -60,9 +60,9 @@ The following hyperparameters were used during training:
|
|
| 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.
|
| 64 |
-
| 0.
|
| 65 |
-
| 0.
|
| 66 |
|
| 67 |
|
| 68 |
### Framework versions
|
|
|
|
| 15 |
|
| 16 |
This model was trained from scratch on the None dataset.
|
| 17 |
It achieves the following results on the evaluation set:
|
| 18 |
+
- Loss: 1.0594
|
| 19 |
+
- Rewards/chosen: -5.1568
|
| 20 |
+
- Rewards/rejected: -5.4496
|
| 21 |
+
- Rewards/accuracies: 0.5625
|
| 22 |
+
- Rewards/margins: 0.2928
|
| 23 |
+
- Logps/rejected: -917.5723
|
| 24 |
+
- Logps/chosen: -895.0606
|
| 25 |
+
- Logits/rejected: -4.8109
|
| 26 |
+
- Logits/chosen: -4.6185
|
| 27 |
|
| 28 |
## Model description
|
| 29 |
|
|
|
|
| 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 |
+
| 0.5292 | 0.26 | 100 | 0.6958 | -1.2149 | -1.3146 | 0.5312 | 0.0997 | -504.0709 | -500.8688 | -3.0075 | -2.9813 |
|
| 64 |
+
| 0.3733 | 0.53 | 200 | 0.9588 | -4.2477 | -4.5339 | 0.5273 | 0.2862 | -826.0007 | -804.1498 | -4.5463 | -4.4070 |
|
| 65 |
+
| 0.3689 | 0.79 | 300 | 1.0594 | -5.1568 | -5.4496 | 0.5625 | 0.2928 | -917.5723 | -895.0606 | -4.8109 | -4.6185 |
|
| 66 |
|
| 67 |
|
| 68 |
### 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": 8.
|
| 7 |
"train_steps_per_second": 0.065
|
| 8 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"epoch": 1.0,
|
| 3 |
+
"train_loss": 0.4399322837512537,
|
| 4 |
+
"train_runtime": 5833.2156,
|
| 5 |
+
"train_samples": 48530,
|
| 6 |
+
"train_samples_per_second": 8.32,
|
| 7 |
"train_steps_per_second": 0.065
|
| 8 |
}
|
config.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"_name_or_path": "/mnt/bn/xuruijie-llm/checkpoints/
|
| 3 |
"architectures": [
|
| 4 |
"MistralForCausalLM"
|
| 5 |
],
|
|
@@ -20,7 +20,7 @@
|
|
| 20 |
"sliding_window": 4096,
|
| 21 |
"tie_word_embeddings": false,
|
| 22 |
"torch_dtype": "bfloat16",
|
| 23 |
-
"transformers_version": "4.
|
| 24 |
"use_cache": false,
|
| 25 |
"vocab_size": 32002
|
| 26 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"_name_or_path": "/mnt/bn/xuruijie-llm/checkpoints/hh-rlhf/sft_0521/checkpoint-5500/",
|
| 3 |
"architectures": [
|
| 4 |
"MistralForCausalLM"
|
| 5 |
],
|
|
|
|
| 20 |
"sliding_window": 4096,
|
| 21 |
"tie_word_embeddings": false,
|
| 22 |
"torch_dtype": "bfloat16",
|
| 23 |
+
"transformers_version": "4.38.2",
|
| 24 |
"use_cache": false,
|
| 25 |
"vocab_size": 32002
|
| 26 |
}
|
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:fdc060d7c5cba351eb050605badb077cae80ee41348189754f56e91a13bf4c06
|
| 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:dae2ed2013519d5a8e0064d100d4202f05d63cb1f9af0f17f77c58e7d46478f2
|
| 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 sha256:6dad5acbe5dbb357fe68df873176da530b0380b90dcff8829bc2e871ca38f1ca
|
| 3 |
size 4540532728
|
runs/May29_00-49-01_n136-129-074/events.out.tfevents.1716915281.n136-129-074.1758650.0
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 sha256:1fd1155f0a88481d8304d980efec6704746371079da9f31a8c9518e47edc8725
|
| 3 |
+
size 59434
|
tokenizer.json
CHANGED
|
@@ -152,7 +152,6 @@
|
|
| 152 |
"end_of_word_suffix": null,
|
| 153 |
"fuse_unk": true,
|
| 154 |
"byte_fallback": true,
|
| 155 |
-
"ignore_merges": false,
|
| 156 |
"vocab": {
|
| 157 |
"<unk>": 0,
|
| 158 |
"<s>": 1,
|
|
|
|
| 152 |
"end_of_word_suffix": null,
|
| 153 |
"fuse_unk": true,
|
| 154 |
"byte_fallback": true,
|
|
|
|
| 155 |
"vocab": {
|
| 156 |
"<unk>": 0,
|
| 157 |
"<s>": 1,
|
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": 8.
|
| 7 |
"train_steps_per_second": 0.065
|
| 8 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"epoch": 1.0,
|
| 3 |
+
"train_loss": 0.4399322837512537,
|
| 4 |
+
"train_runtime": 5833.2156,
|
| 5 |
+
"train_samples": 48530,
|
| 6 |
+
"train_samples_per_second": 8.32,
|
| 7 |
"train_steps_per_second": 0.065
|
| 8 |
}
|
trainer_state.json
CHANGED
|
@@ -1,22 +1,22 @@
|
|
| 1 |
{
|
| 2 |
"best_metric": null,
|
| 3 |
"best_model_checkpoint": null,
|
| 4 |
-
"epoch":
|
| 5 |
"eval_steps": 100,
|
| 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": 0.0,
|
| 13 |
-
"grad_norm":
|
| 14 |
-
"learning_rate": 1.
|
| 15 |
-
"logits/chosen": -2.
|
| 16 |
-
"logits/rejected": -2.
|
| 17 |
-
"logps/chosen": -
|
| 18 |
-
"logps/rejected": -
|
| 19 |
-
"loss": 0.
|
| 20 |
"rewards/accuracies": 0.0,
|
| 21 |
"rewards/chosen": 0.0,
|
| 22 |
"rewards/margins": 0.0,
|
|
@@ -24,650 +24,1190 @@
|
|
| 24 |
"step": 1
|
| 25 |
},
|
| 26 |
{
|
| 27 |
-
"epoch": 0.
|
| 28 |
-
"grad_norm":
|
| 29 |
-
"learning_rate":
|
| 30 |
-
"logits/chosen": -2.
|
| 31 |
-
"logits/rejected": -2.
|
| 32 |
-
"logps/chosen": -
|
| 33 |
-
"logps/rejected": -
|
| 34 |
"loss": 0.6931,
|
| 35 |
-
"rewards/accuracies": 0.
|
| 36 |
-
"rewards/chosen": 0.
|
| 37 |
-
"rewards/margins": 0.
|
| 38 |
-
"rewards/rejected": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
"step": 10
|
| 40 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
{
|
| 42 |
"epoch": 0.05,
|
| 43 |
-
"grad_norm":
|
| 44 |
-
"learning_rate": 2.
|
| 45 |
-
"logits/chosen": -2.
|
| 46 |
-
"logits/rejected": -2.
|
| 47 |
-
"logps/chosen": -
|
| 48 |
-
"logps/rejected": -
|
| 49 |
-
"loss": 0.
|
| 50 |
-
"rewards/accuracies": 0.
|
| 51 |
-
"rewards/chosen": 0.
|
| 52 |
-
"rewards/margins": 0.
|
| 53 |
-
"rewards/rejected": 0.
|
| 54 |
"step": 20
|
| 55 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 56 |
{
|
| 57 |
"epoch": 0.08,
|
| 58 |
-
"grad_norm":
|
| 59 |
-
"learning_rate": 3.
|
| 60 |
-
"logits/chosen": -2.
|
| 61 |
-
"logits/rejected": -2.
|
| 62 |
-
"logps/chosen": -
|
| 63 |
-
"logps/rejected": -
|
| 64 |
-
"loss": 0.
|
| 65 |
-
"rewards/accuracies": 0.
|
| 66 |
-
"rewards/chosen": 0.
|
| 67 |
-
"rewards/margins": 0.
|
| 68 |
-
"rewards/rejected": 0.
|
| 69 |
"step": 30
|
| 70 |
},
|
| 71 |
{
|
| 72 |
-
"epoch": 0.
|
| 73 |
-
"grad_norm":
|
| 74 |
-
"learning_rate":
|
| 75 |
-
"logits/chosen": -2.
|
| 76 |
-
"logits/rejected": -2.
|
| 77 |
-
"logps/chosen": -
|
| 78 |
-
"logps/rejected": -
|
| 79 |
-
"loss": 0.
|
| 80 |
-
"rewards/accuracies": 0.
|
| 81 |
-
"rewards/chosen": 0.
|
| 82 |
-
"rewards/margins": 0.
|
| 83 |
-
"rewards/rejected": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 84 |
"step": 40
|
| 85 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 86 |
{
|
| 87 |
"epoch": 0.13,
|
| 88 |
-
"grad_norm":
|
| 89 |
-
"learning_rate": 4.
|
| 90 |
-
"logits/chosen": -2.
|
| 91 |
-
"logits/rejected": -2.
|
| 92 |
-
"logps/chosen": -
|
| 93 |
-
"logps/rejected": -
|
| 94 |
-
"loss": 0.
|
| 95 |
-
"rewards/accuracies": 0.
|
| 96 |
-
"rewards/chosen": 0.
|
| 97 |
-
"rewards/margins": 0.
|
| 98 |
-
"rewards/rejected": 0.
|
| 99 |
"step": 50
|
| 100 |
},
|
| 101 |
{
|
| 102 |
-
"epoch": 0.
|
| 103 |
-
"grad_norm":
|
| 104 |
-
"learning_rate": 4.
|
| 105 |
-
"logits/chosen": -2.
|
| 106 |
-
"logits/rejected": -2.
|
| 107 |
-
"logps/chosen": -
|
| 108 |
-
"logps/rejected": -
|
| 109 |
-
"loss": 0.
|
| 110 |
-
"rewards/accuracies": 0.
|
| 111 |
-
"rewards/chosen": 0.
|
| 112 |
-
"rewards/margins": 0.
|
| 113 |
-
"rewards/rejected":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 114 |
"step": 60
|
| 115 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 116 |
{
|
| 117 |
"epoch": 0.18,
|
| 118 |
-
"grad_norm":
|
| 119 |
-
"learning_rate": 4.
|
| 120 |
-
"logits/chosen": -2.
|
| 121 |
-
"logits/rejected": -2.
|
| 122 |
-
"logps/chosen": -
|
| 123 |
-
"logps/rejected": -
|
| 124 |
-
"loss": 0.
|
| 125 |
-
"rewards/accuracies": 0.
|
| 126 |
-
"rewards/chosen":
|
| 127 |
-
"rewards/margins": 0.
|
| 128 |
-
"rewards/rejected":
|
| 129 |
"step": 70
|
| 130 |
},
|
| 131 |
{
|
| 132 |
"epoch": 0.2,
|
| 133 |
-
"grad_norm":
|
| 134 |
-
"learning_rate": 4.
|
| 135 |
-
"logits/chosen": -2.
|
| 136 |
-
"logits/rejected": -2.
|
| 137 |
-
"logps/chosen": -
|
| 138 |
-
"logps/rejected": -
|
| 139 |
-
"loss": 0.
|
| 140 |
-
"rewards/accuracies": 0.
|
| 141 |
-
"rewards/chosen":
|
| 142 |
-
"rewards/margins": 0.
|
| 143 |
-
"rewards/rejected":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 144 |
"step": 80
|
| 145 |
},
|
| 146 |
{
|
| 147 |
-
"epoch": 0.
|
| 148 |
-
"grad_norm":
|
| 149 |
-
"learning_rate": 4.
|
| 150 |
-
"logits/chosen": -2.
|
| 151 |
-
"logits/rejected": -2.
|
| 152 |
-
"logps/chosen": -
|
| 153 |
-
"logps/rejected": -
|
| 154 |
-
"loss": 0.
|
| 155 |
-
"rewards/accuracies": 0.
|
| 156 |
-
"rewards/chosen":
|
| 157 |
-
"rewards/margins": 0.
|
| 158 |
-
"rewards/rejected": -
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 159 |
"step": 90
|
| 160 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 161 |
{
|
| 162 |
"epoch": 0.26,
|
| 163 |
-
"grad_norm":
|
| 164 |
-
"learning_rate": 4.
|
| 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": 0.26,
|
| 178 |
-
"eval_logits/chosen": -2.
|
| 179 |
-
"eval_logits/rejected": -
|
| 180 |
-
"eval_logps/chosen": -
|
| 181 |
-
"eval_logps/rejected": -
|
| 182 |
-
"eval_loss": 0.
|
| 183 |
-
"eval_rewards/accuracies": 0.
|
| 184 |
-
"eval_rewards/chosen": -
|
| 185 |
-
"eval_rewards/margins": 0.
|
| 186 |
-
"eval_rewards/rejected": -
|
| 187 |
-
"eval_runtime": 97.
|
| 188 |
-
"eval_samples_per_second": 20.
|
| 189 |
-
"eval_steps_per_second": 0.
|
| 190 |
"step": 100
|
| 191 |
},
|
| 192 |
{
|
| 193 |
"epoch": 0.28,
|
| 194 |
-
"grad_norm":
|
| 195 |
-
"learning_rate": 4.
|
| 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": 4.
|
| 211 |
-
"logits/chosen": -
|
| 212 |
-
"logits/rejected": -
|
| 213 |
-
"logps/chosen": -
|
| 214 |
-
"logps/rejected": -
|
| 215 |
-
"loss": 0.
|
| 216 |
"rewards/accuracies": 0.824999988079071,
|
| 217 |
-
"rewards/chosen": -
|
| 218 |
-
"rewards/margins":
|
| 219 |
-
"rewards/rejected": -
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 220 |
"step": 120
|
| 221 |
},
|
| 222 |
{
|
| 223 |
"epoch": 0.33,
|
| 224 |
-
"grad_norm":
|
| 225 |
-
"learning_rate": 4.
|
| 226 |
-
"logits/chosen": -3.
|
| 227 |
-
"logits/rejected": -3.
|
| 228 |
-
"logps/chosen": -
|
| 229 |
-
"logps/rejected": -
|
| 230 |
-
"loss": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 231 |
"rewards/accuracies": 0.8125,
|
| 232 |
-
"rewards/chosen": -
|
| 233 |
-
"rewards/margins":
|
| 234 |
-
"rewards/rejected": -
|
| 235 |
"step": 130
|
| 236 |
},
|
| 237 |
{
|
| 238 |
"epoch": 0.36,
|
| 239 |
-
"grad_norm":
|
| 240 |
-
"learning_rate": 4.
|
| 241 |
-
"logits/chosen": -3.
|
| 242 |
-
"logits/rejected": -3.
|
| 243 |
-
"logps/chosen": -392.
|
| 244 |
-
"logps/rejected": -
|
| 245 |
-
"loss": 0.
|
| 246 |
-
"rewards/accuracies": 0.
|
| 247 |
-
"rewards/chosen": -1.
|
| 248 |
-
"rewards/margins": 1.
|
| 249 |
-
"rewards/rejected": -2.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 250 |
"step": 140
|
| 251 |
},
|
| 252 |
{
|
| 253 |
"epoch": 0.38,
|
| 254 |
-
"grad_norm":
|
| 255 |
-
"learning_rate": 3.
|
| 256 |
-
"logits/chosen": -3.
|
| 257 |
-
"logits/rejected": -3.
|
| 258 |
-
"logps/chosen": -
|
| 259 |
-
"logps/rejected": -
|
| 260 |
-
"loss": 0.
|
| 261 |
-
"rewards/accuracies": 0.
|
| 262 |
-
"rewards/chosen": -1.
|
| 263 |
-
"rewards/margins":
|
| 264 |
-
"rewards/rejected": -
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 265 |
"step": 150
|
| 266 |
},
|
| 267 |
{
|
| 268 |
"epoch": 0.41,
|
| 269 |
-
"grad_norm":
|
| 270 |
-
"learning_rate": 3.
|
| 271 |
-
"logits/chosen": -3.
|
| 272 |
-
"logits/rejected": -
|
| 273 |
-
"logps/chosen": -
|
| 274 |
-
"logps/rejected": -
|
| 275 |
-
"loss": 0.
|
| 276 |
-
"rewards/accuracies": 0.
|
| 277 |
-
"rewards/chosen": -1.
|
| 278 |
-
"rewards/margins": 1.
|
| 279 |
-
"rewards/rejected": -
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 280 |
"step": 160
|
| 281 |
},
|
| 282 |
{
|
| 283 |
"epoch": 0.43,
|
| 284 |
-
"grad_norm":
|
| 285 |
-
"learning_rate": 3.
|
| 286 |
-
"logits/chosen": -
|
| 287 |
-
"logits/rejected": -
|
| 288 |
-
"logps/chosen": -
|
| 289 |
-
"logps/rejected": -
|
| 290 |
-
"loss": 0.
|
| 291 |
-
"rewards/accuracies": 0.
|
| 292 |
-
"rewards/chosen": -1.
|
| 293 |
-
"rewards/margins": 1.
|
| 294 |
-
"rewards/rejected": -
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 295 |
"step": 170
|
| 296 |
},
|
| 297 |
{
|
| 298 |
"epoch": 0.46,
|
| 299 |
-
"grad_norm":
|
| 300 |
-
"learning_rate": 3.
|
| 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": -3.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 310 |
"step": 180
|
| 311 |
},
|
| 312 |
{
|
| 313 |
"epoch": 0.49,
|
| 314 |
-
"grad_norm":
|
| 315 |
-
"learning_rate": 3.
|
| 316 |
-
"logits/chosen": -4.
|
| 317 |
-
"logits/rejected": -4.
|
| 318 |
-
"logps/chosen": -
|
| 319 |
-
"logps/rejected": -
|
| 320 |
-
"loss": 0.
|
| 321 |
-
"rewards/accuracies": 0.
|
| 322 |
-
"rewards/chosen": -
|
| 323 |
-
"rewards/margins":
|
| 324 |
-
"rewards/rejected": -
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 325 |
"step": 190
|
| 326 |
},
|
| 327 |
{
|
| 328 |
"epoch": 0.51,
|
| 329 |
-
"grad_norm":
|
| 330 |
-
"learning_rate": 2.
|
| 331 |
-
"logits/chosen": -
|
| 332 |
-
"logits/rejected": -4.
|
| 333 |
-
"logps/chosen": -
|
| 334 |
-
"logps/rejected": -
|
| 335 |
-
"loss": 0.
|
| 336 |
-
"rewards/accuracies": 0.
|
| 337 |
-
"rewards/chosen": -
|
| 338 |
-
"rewards/margins":
|
| 339 |
-
"rewards/rejected": -3.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 340 |
"step": 200
|
| 341 |
},
|
| 342 |
{
|
| 343 |
-
"epoch": 0.
|
| 344 |
-
"eval_logits/chosen": -
|
| 345 |
-
"eval_logits/rejected": -
|
| 346 |
-
"eval_logps/chosen": -
|
| 347 |
-
"eval_logps/rejected": -
|
| 348 |
-
"eval_loss":
|
| 349 |
-
"eval_rewards/accuracies": 0.
|
| 350 |
-
"eval_rewards/chosen": -4.
|
| 351 |
-
"eval_rewards/margins": 0.
|
| 352 |
-
"eval_rewards/rejected": -4.
|
| 353 |
-
"eval_runtime": 97.
|
| 354 |
-
"eval_samples_per_second": 20.
|
| 355 |
-
"eval_steps_per_second": 0.
|
| 356 |
"step": 200
|
| 357 |
},
|
| 358 |
{
|
| 359 |
"epoch": 0.54,
|
| 360 |
-
"grad_norm":
|
| 361 |
-
"learning_rate": 2.
|
| 362 |
-
"logits/chosen": -
|
| 363 |
-
"logits/rejected": -4.
|
| 364 |
-
"logps/chosen": -
|
| 365 |
-
"logps/rejected": -
|
| 366 |
-
"loss": 0.
|
| 367 |
-
"rewards/accuracies": 0.
|
| 368 |
-
"rewards/chosen": -1.
|
| 369 |
-
"rewards/margins":
|
| 370 |
-
"rewards/rejected": -
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 371 |
"step": 210
|
| 372 |
},
|
| 373 |
{
|
| 374 |
-
"epoch": 0.
|
| 375 |
-
"grad_norm":
|
| 376 |
-
"learning_rate": 2.
|
| 377 |
-
"logits/chosen": -
|
| 378 |
-
"logits/rejected": -4.
|
| 379 |
-
"logps/chosen": -
|
| 380 |
-
"logps/rejected": -
|
| 381 |
-
"loss": 0.
|
| 382 |
-
"rewards/accuracies": 0.
|
| 383 |
-
"rewards/chosen": -
|
| 384 |
-
"rewards/margins":
|
| 385 |
-
"rewards/rejected": -
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 386 |
"step": 220
|
| 387 |
},
|
| 388 |
{
|
| 389 |
"epoch": 0.59,
|
| 390 |
-
"grad_norm":
|
| 391 |
-
"learning_rate": 2.
|
| 392 |
-
"logits/chosen": -
|
| 393 |
-
"logits/rejected": -4.
|
| 394 |
-
"logps/chosen": -
|
| 395 |
-
"logps/rejected": -
|
| 396 |
-
"loss": 0.
|
| 397 |
"rewards/accuracies": 0.862500011920929,
|
| 398 |
-
"rewards/chosen": -
|
| 399 |
-
"rewards/margins":
|
| 400 |
-
"rewards/rejected": -
|
| 401 |
-
"step":
|
| 402 |
},
|
| 403 |
{
|
| 404 |
"epoch": 0.61,
|
| 405 |
-
"grad_norm":
|
| 406 |
-
"learning_rate":
|
| 407 |
-
"logits/chosen": -4.
|
| 408 |
-
"logits/rejected": -
|
| 409 |
-
"logps/chosen": -
|
| 410 |
-
"logps/rejected": -
|
| 411 |
-
"loss": 0.
|
| 412 |
-
"rewards/accuracies": 0.
|
| 413 |
-
"rewards/chosen": -2.
|
| 414 |
-
"rewards/margins": 2.
|
| 415 |
-
"rewards/rejected": -
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 416 |
"step": 240
|
| 417 |
},
|
| 418 |
{
|
| 419 |
-
"epoch": 0.
|
| 420 |
-
"grad_norm":
|
| 421 |
-
"learning_rate": 1.
|
| 422 |
-
"logits/chosen": -4.
|
| 423 |
-
"logits/rejected": -4.
|
| 424 |
-
"logps/chosen": -
|
| 425 |
-
"logps/rejected": -
|
| 426 |
-
"loss": 0.
|
| 427 |
-
"rewards/accuracies": 0.
|
| 428 |
-
"rewards/chosen": -
|
| 429 |
-
"rewards/margins": 2.
|
| 430 |
-
"rewards/rejected": -4.
|
| 431 |
-
"step":
|
| 432 |
},
|
| 433 |
{
|
| 434 |
"epoch": 0.66,
|
| 435 |
-
"grad_norm":
|
| 436 |
-
"learning_rate": 1.
|
| 437 |
-
"logits/chosen": -4.
|
| 438 |
-
"logits/rejected": -4.
|
| 439 |
-
"logps/chosen": -
|
| 440 |
-
"logps/rejected": -
|
| 441 |
-
"loss": 0.
|
| 442 |
-
"rewards/accuracies": 0.
|
| 443 |
-
"rewards/chosen": -1.
|
| 444 |
-
"rewards/margins":
|
| 445 |
-
"rewards/rejected": -
|
| 446 |
-
"step":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 447 |
},
|
| 448 |
{
|
| 449 |
"epoch": 0.69,
|
| 450 |
-
"grad_norm":
|
| 451 |
-
"learning_rate": 1.
|
| 452 |
-
"logits/chosen": -4.
|
| 453 |
-
"logits/rejected": -4.
|
| 454 |
-
"logps/chosen": -
|
| 455 |
-
"logps/rejected": -
|
| 456 |
-
"loss": 0.
|
| 457 |
-
"rewards/accuracies": 0.
|
| 458 |
-
"rewards/chosen": -2.
|
| 459 |
-
"rewards/margins": 2.
|
| 460 |
-
"rewards/rejected": -4.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 461 |
"step": 270
|
| 462 |
},
|
| 463 |
{
|
| 464 |
"epoch": 0.72,
|
| 465 |
-
"grad_norm":
|
| 466 |
-
"learning_rate": 1.
|
| 467 |
-
"logits/chosen": -4.
|
| 468 |
-
"logits/rejected": -4.
|
| 469 |
-
"logps/chosen": -
|
| 470 |
-
"logps/rejected": -
|
| 471 |
-
"loss": 0.
|
| 472 |
-
"rewards/accuracies": 0.
|
| 473 |
-
"rewards/chosen": -
|
| 474 |
-
"rewards/margins":
|
| 475 |
-
"rewards/rejected": -
|
| 476 |
-
"step":
|
| 477 |
},
|
| 478 |
{
|
| 479 |
"epoch": 0.74,
|
| 480 |
-
"grad_norm":
|
| 481 |
-
"learning_rate": 9.
|
| 482 |
-
"logits/chosen": -4.
|
| 483 |
-
"logits/rejected": -4.
|
| 484 |
-
"logps/chosen": -
|
| 485 |
-
"logps/rejected": -
|
| 486 |
-
"loss": 0.
|
| 487 |
-
"rewards/accuracies": 0.
|
| 488 |
-
"rewards/chosen": -
|
| 489 |
-
"rewards/margins":
|
| 490 |
-
"rewards/rejected": -3.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 491 |
"step": 290
|
| 492 |
},
|
| 493 |
{
|
| 494 |
-
"epoch": 0.
|
| 495 |
-
"grad_norm":
|
| 496 |
-
"learning_rate": 7.
|
| 497 |
-
"logits/chosen": -4.
|
| 498 |
-
"logits/rejected": -
|
| 499 |
-
"logps/chosen": -
|
| 500 |
-
"logps/rejected": -
|
| 501 |
-
"loss": 0.
|
| 502 |
-
"rewards/accuracies": 0.
|
| 503 |
-
"rewards/chosen": -1.
|
| 504 |
-
"rewards/margins":
|
| 505 |
-
"rewards/rejected": -
|
| 506 |
-
"step":
|
| 507 |
},
|
| 508 |
{
|
| 509 |
-
"epoch": 0.
|
| 510 |
-
"
|
| 511 |
-
"
|
| 512 |
-
"
|
| 513 |
-
"
|
| 514 |
-
"
|
| 515 |
-
"
|
| 516 |
-
"
|
| 517 |
-
"
|
| 518 |
-
"
|
| 519 |
-
"
|
| 520 |
-
"
|
| 521 |
-
"eval_steps_per_second": 0.326,
|
| 522 |
"step": 300
|
| 523 |
},
|
| 524 |
{
|
| 525 |
"epoch": 0.79,
|
| 526 |
-
"
|
| 527 |
-
"
|
| 528 |
-
"
|
| 529 |
-
"
|
| 530 |
-
"
|
| 531 |
-
"
|
| 532 |
-
"
|
| 533 |
-
"
|
| 534 |
-
"
|
| 535 |
-
"
|
| 536 |
-
"
|
| 537 |
-
"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 538 |
},
|
| 539 |
{
|
| 540 |
"epoch": 0.82,
|
| 541 |
-
"grad_norm":
|
| 542 |
-
"learning_rate": 4.
|
| 543 |
-
"logits/chosen": -4.
|
| 544 |
-
"logits/rejected": -
|
| 545 |
-
"logps/chosen": -
|
| 546 |
-
"logps/rejected": -
|
| 547 |
-
"loss": 0.
|
| 548 |
-
"rewards/accuracies": 0.
|
| 549 |
-
"rewards/chosen": -2.
|
| 550 |
-
"rewards/margins":
|
| 551 |
-
"rewards/rejected": -4.
|
| 552 |
-
"step":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 553 |
},
|
| 554 |
{
|
| 555 |
"epoch": 0.84,
|
| 556 |
-
"grad_norm":
|
| 557 |
-
"learning_rate": 3.
|
| 558 |
-
"logits/chosen": -4.
|
| 559 |
-
"logits/rejected": -4.
|
| 560 |
-
"logps/chosen": -
|
| 561 |
-
"logps/rejected": -
|
| 562 |
-
"loss": 0.
|
| 563 |
-
"rewards/accuracies": 0.
|
| 564 |
-
"rewards/chosen": -2.
|
| 565 |
-
"rewards/margins":
|
| 566 |
-
"rewards/rejected": -
|
| 567 |
-
"step":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 568 |
},
|
| 569 |
{
|
| 570 |
"epoch": 0.87,
|
| 571 |
-
"grad_norm":
|
| 572 |
-
"learning_rate": 2.
|
| 573 |
-
"logits/chosen": -4.
|
| 574 |
-
"logits/rejected": -
|
| 575 |
-
"logps/chosen": -
|
| 576 |
-
"logps/rejected": -
|
| 577 |
-
"loss": 0.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 578 |
"rewards/accuracies": 0.8374999761581421,
|
| 579 |
-
"rewards/chosen": -
|
| 580 |
-
"rewards/margins":
|
| 581 |
-
"rewards/rejected": -
|
| 582 |
"step": 340
|
| 583 |
},
|
| 584 |
{
|
| 585 |
-
"epoch": 0.
|
| 586 |
-
"grad_norm":
|
| 587 |
-
"learning_rate": 1.
|
| 588 |
-
"logits/chosen": -4.
|
| 589 |
-
"logits/rejected": -4.
|
| 590 |
-
"logps/chosen": -
|
| 591 |
-
"logps/rejected": -
|
| 592 |
-
"loss": 0.
|
| 593 |
-
"rewards/accuracies": 0.
|
| 594 |
-
"rewards/chosen": -
|
| 595 |
-
"rewards/margins": 2.
|
| 596 |
-
"rewards/rejected": -4.
|
| 597 |
-
"step":
|
| 598 |
},
|
| 599 |
{
|
| 600 |
"epoch": 0.92,
|
| 601 |
-
"grad_norm":
|
| 602 |
-
"learning_rate":
|
| 603 |
-
"logits/chosen": -4.
|
| 604 |
-
"logits/rejected": -
|
| 605 |
-
"logps/chosen": -
|
| 606 |
-
"logps/rejected": -
|
| 607 |
-
"loss": 0.
|
| 608 |
-
"rewards/accuracies": 0.
|
| 609 |
-
"rewards/chosen": -2.
|
| 610 |
-
"rewards/margins":
|
| 611 |
-
"rewards/rejected": -4.
|
| 612 |
-
"step":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 613 |
},
|
| 614 |
{
|
| 615 |
"epoch": 0.95,
|
| 616 |
-
"grad_norm":
|
| 617 |
-
"learning_rate":
|
| 618 |
-
"logits/chosen": -4.
|
| 619 |
-
"logits/rejected": -4.
|
| 620 |
-
"logps/chosen": -
|
| 621 |
-
"logps/rejected": -
|
| 622 |
-
"loss": 0.
|
| 623 |
-
"rewards/accuracies": 0.
|
| 624 |
-
"rewards/chosen": -
|
| 625 |
-
"rewards/margins":
|
| 626 |
-
"rewards/rejected": -4.
|
| 627 |
-
"step":
|
| 628 |
},
|
| 629 |
{
|
| 630 |
-
"epoch": 0.
|
| 631 |
-
"grad_norm":
|
| 632 |
-
"learning_rate":
|
| 633 |
-
"logits/chosen": -4.
|
| 634 |
-
"logits/rejected": -
|
| 635 |
-
"logps/chosen": -
|
| 636 |
-
"logps/rejected": -
|
| 637 |
-
"loss": 0.
|
| 638 |
-
"rewards/accuracies": 0.
|
| 639 |
-
"rewards/chosen": -1.
|
| 640 |
-
"rewards/margins":
|
| 641 |
-
"rewards/rejected": -
|
| 642 |
-
"step":
|
| 643 |
},
|
| 644 |
{
|
| 645 |
-
"epoch":
|
| 646 |
-
"grad_norm":
|
| 647 |
-
"learning_rate":
|
| 648 |
-
"logits/chosen": -4.
|
| 649 |
-
"logits/rejected": -
|
| 650 |
-
"logps/chosen": -
|
| 651 |
-
"logps/rejected": -
|
| 652 |
-
"loss": 0.
|
| 653 |
"rewards/accuracies": 0.862500011920929,
|
| 654 |
-
"rewards/chosen": -
|
| 655 |
-
"rewards/margins": 2.
|
| 656 |
-
"rewards/rejected": -4.
|
| 657 |
-
"step":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 658 |
},
|
| 659 |
{
|
| 660 |
"epoch": 1.0,
|
| 661 |
-
"step":
|
| 662 |
"total_flos": 0.0,
|
| 663 |
-
"train_loss": 0.
|
| 664 |
-
"train_runtime":
|
| 665 |
-
"train_samples_per_second": 8.
|
| 666 |
"train_steps_per_second": 0.065
|
| 667 |
}
|
| 668 |
],
|
| 669 |
-
"logging_steps":
|
| 670 |
-
"max_steps":
|
| 671 |
"num_input_tokens_seen": 0,
|
| 672 |
"num_train_epochs": 1,
|
| 673 |
"save_steps": 100,
|
|
|
|
| 1 |
{
|
| 2 |
"best_metric": null,
|
| 3 |
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 0.9986824769433466,
|
| 5 |
"eval_steps": 100,
|
| 6 |
+
"global_step": 379,
|
| 7 |
"is_hyper_param_search": false,
|
| 8 |
"is_local_process_zero": true,
|
| 9 |
"is_world_process_zero": true,
|
| 10 |
"log_history": [
|
| 11 |
{
|
| 12 |
"epoch": 0.0,
|
| 13 |
+
"grad_norm": 32.10821322863283,
|
| 14 |
+
"learning_rate": 1.3157894736842104e-08,
|
| 15 |
+
"logits/chosen": -2.219799041748047,
|
| 16 |
+
"logits/rejected": -2.229109525680542,
|
| 17 |
+
"logps/chosen": -269.856201171875,
|
| 18 |
+
"logps/rejected": -192.3697509765625,
|
| 19 |
+
"loss": 0.6931,
|
| 20 |
"rewards/accuracies": 0.0,
|
| 21 |
"rewards/chosen": 0.0,
|
| 22 |
"rewards/margins": 0.0,
|
|
|
|
| 24 |
"step": 1
|
| 25 |
},
|
| 26 |
{
|
| 27 |
+
"epoch": 0.01,
|
| 28 |
+
"grad_norm": 35.006735982686784,
|
| 29 |
+
"learning_rate": 6.578947368421052e-08,
|
| 30 |
+
"logits/chosen": -2.276287078857422,
|
| 31 |
+
"logits/rejected": -2.2080204486846924,
|
| 32 |
+
"logps/chosen": -283.5397644042969,
|
| 33 |
+
"logps/rejected": -208.59442138671875,
|
| 34 |
"loss": 0.6931,
|
| 35 |
+
"rewards/accuracies": 0.40625,
|
| 36 |
+
"rewards/chosen": 0.0013411559630185366,
|
| 37 |
+
"rewards/margins": 0.00017823810048867017,
|
| 38 |
+
"rewards/rejected": 0.0011629178188741207,
|
| 39 |
+
"step": 5
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.03,
|
| 43 |
+
"grad_norm": 36.439618208906154,
|
| 44 |
+
"learning_rate": 1.3157894736842104e-07,
|
| 45 |
+
"logits/chosen": -2.3274893760681152,
|
| 46 |
+
"logits/rejected": -2.2322466373443604,
|
| 47 |
+
"logps/chosen": -286.6865539550781,
|
| 48 |
+
"logps/rejected": -206.5416717529297,
|
| 49 |
+
"loss": 0.6926,
|
| 50 |
+
"rewards/accuracies": 0.48750001192092896,
|
| 51 |
+
"rewards/chosen": 0.00261278566904366,
|
| 52 |
+
"rewards/margins": -3.606556128943339e-05,
|
| 53 |
+
"rewards/rejected": 0.0026488511357456446,
|
| 54 |
"step": 10
|
| 55 |
},
|
| 56 |
+
{
|
| 57 |
+
"epoch": 0.04,
|
| 58 |
+
"grad_norm": 34.873528802219326,
|
| 59 |
+
"learning_rate": 1.9736842105263157e-07,
|
| 60 |
+
"logits/chosen": -2.3293230533599854,
|
| 61 |
+
"logits/rejected": -2.2996506690979004,
|
| 62 |
+
"logps/chosen": -289.71771240234375,
|
| 63 |
+
"logps/rejected": -237.6280059814453,
|
| 64 |
+
"loss": 0.6906,
|
| 65 |
+
"rewards/accuracies": 0.612500011920929,
|
| 66 |
+
"rewards/chosen": 0.017419874668121338,
|
| 67 |
+
"rewards/margins": 0.004653422627598047,
|
| 68 |
+
"rewards/rejected": 0.012766450643539429,
|
| 69 |
+
"step": 15
|
| 70 |
+
},
|
| 71 |
{
|
| 72 |
"epoch": 0.05,
|
| 73 |
+
"grad_norm": 31.898185632963774,
|
| 74 |
+
"learning_rate": 2.631578947368421e-07,
|
| 75 |
+
"logits/chosen": -2.3196263313293457,
|
| 76 |
+
"logits/rejected": -2.2592759132385254,
|
| 77 |
+
"logps/chosen": -263.24908447265625,
|
| 78 |
+
"logps/rejected": -202.05458068847656,
|
| 79 |
+
"loss": 0.6857,
|
| 80 |
+
"rewards/accuracies": 0.762499988079071,
|
| 81 |
+
"rewards/chosen": 0.05536944791674614,
|
| 82 |
+
"rewards/margins": 0.01851554773747921,
|
| 83 |
+
"rewards/rejected": 0.03685389831662178,
|
| 84 |
"step": 20
|
| 85 |
},
|
| 86 |
+
{
|
| 87 |
+
"epoch": 0.07,
|
| 88 |
+
"grad_norm": 28.076450357377052,
|
| 89 |
+
"learning_rate": 3.2894736842105264e-07,
|
| 90 |
+
"logits/chosen": -2.3921802043914795,
|
| 91 |
+
"logits/rejected": -2.377410650253296,
|
| 92 |
+
"logps/chosen": -260.9800720214844,
|
| 93 |
+
"logps/rejected": -214.38821411132812,
|
| 94 |
+
"loss": 0.6763,
|
| 95 |
+
"rewards/accuracies": 0.762499988079071,
|
| 96 |
+
"rewards/chosen": 0.13004128634929657,
|
| 97 |
+
"rewards/margins": 0.03864985704421997,
|
| 98 |
+
"rewards/rejected": 0.0913914293050766,
|
| 99 |
+
"step": 25
|
| 100 |
+
},
|
| 101 |
{
|
| 102 |
"epoch": 0.08,
|
| 103 |
+
"grad_norm": 25.341140024963043,
|
| 104 |
+
"learning_rate": 3.9473684210526315e-07,
|
| 105 |
+
"logits/chosen": -2.477496385574341,
|
| 106 |
+
"logits/rejected": -2.506354808807373,
|
| 107 |
+
"logps/chosen": -255.5835723876953,
|
| 108 |
+
"logps/rejected": -223.4532470703125,
|
| 109 |
+
"loss": 0.6664,
|
| 110 |
+
"rewards/accuracies": 0.699999988079071,
|
| 111 |
+
"rewards/chosen": 0.2109464406967163,
|
| 112 |
+
"rewards/margins": 0.04907592386007309,
|
| 113 |
+
"rewards/rejected": 0.16187050938606262,
|
| 114 |
"step": 30
|
| 115 |
},
|
| 116 |
{
|
| 117 |
+
"epoch": 0.09,
|
| 118 |
+
"grad_norm": 23.40327238126855,
|
| 119 |
+
"learning_rate": 4.6052631578947365e-07,
|
| 120 |
+
"logits/chosen": -2.6838037967681885,
|
| 121 |
+
"logits/rejected": -2.6064510345458984,
|
| 122 |
+
"logps/chosen": -272.02825927734375,
|
| 123 |
+
"logps/rejected": -207.57565307617188,
|
| 124 |
+
"loss": 0.6576,
|
| 125 |
+
"rewards/accuracies": 0.800000011920929,
|
| 126 |
+
"rewards/chosen": 0.3294451832771301,
|
| 127 |
+
"rewards/margins": 0.12016526609659195,
|
| 128 |
+
"rewards/rejected": 0.20927992463111877,
|
| 129 |
+
"step": 35
|
| 130 |
+
},
|
| 131 |
+
{
|
| 132 |
+
"epoch": 0.11,
|
| 133 |
+
"grad_norm": 21.272616898577162,
|
| 134 |
+
"learning_rate": 4.999575626062319e-07,
|
| 135 |
+
"logits/chosen": -2.747399091720581,
|
| 136 |
+
"logits/rejected": -2.7062249183654785,
|
| 137 |
+
"logps/chosen": -249.28085327148438,
|
| 138 |
+
"logps/rejected": -202.8720245361328,
|
| 139 |
+
"loss": 0.6504,
|
| 140 |
+
"rewards/accuracies": 0.6625000238418579,
|
| 141 |
+
"rewards/chosen": 0.3287069499492645,
|
| 142 |
+
"rewards/margins": 0.0975252240896225,
|
| 143 |
+
"rewards/rejected": 0.23118171095848083,
|
| 144 |
"step": 40
|
| 145 |
},
|
| 146 |
+
{
|
| 147 |
+
"epoch": 0.12,
|
| 148 |
+
"grad_norm": 18.3841620762551,
|
| 149 |
+
"learning_rate": 4.994803073715569e-07,
|
| 150 |
+
"logits/chosen": -2.788217067718506,
|
| 151 |
+
"logits/rejected": -2.736720561981201,
|
| 152 |
+
"logps/chosen": -260.8193359375,
|
| 153 |
+
"logps/rejected": -202.5912322998047,
|
| 154 |
+
"loss": 0.6444,
|
| 155 |
+
"rewards/accuracies": 0.7875000238418579,
|
| 156 |
+
"rewards/chosen": 0.4205988049507141,
|
| 157 |
+
"rewards/margins": 0.14394915103912354,
|
| 158 |
+
"rewards/rejected": 0.2766496539115906,
|
| 159 |
+
"step": 45
|
| 160 |
+
},
|
| 161 |
{
|
| 162 |
"epoch": 0.13,
|
| 163 |
+
"grad_norm": 18.656759383093647,
|
| 164 |
+
"learning_rate": 4.984737660598186e-07,
|
| 165 |
+
"logits/chosen": -2.8456408977508545,
|
| 166 |
+
"logits/rejected": -2.7947795391082764,
|
| 167 |
+
"logps/chosen": -223.0256805419922,
|
| 168 |
+
"logps/rejected": -197.88247680664062,
|
| 169 |
+
"loss": 0.6469,
|
| 170 |
+
"rewards/accuracies": 0.675000011920929,
|
| 171 |
+
"rewards/chosen": 0.3571344316005707,
|
| 172 |
+
"rewards/margins": 0.09834496676921844,
|
| 173 |
+
"rewards/rejected": 0.2587894797325134,
|
| 174 |
"step": 50
|
| 175 |
},
|
| 176 |
{
|
| 177 |
+
"epoch": 0.14,
|
| 178 |
+
"grad_norm": 20.93264303618464,
|
| 179 |
+
"learning_rate": 4.969400741032999e-07,
|
| 180 |
+
"logits/chosen": -2.8368711471557617,
|
| 181 |
+
"logits/rejected": -2.8192451000213623,
|
| 182 |
+
"logps/chosen": -234.53515625,
|
| 183 |
+
"logps/rejected": -199.7665252685547,
|
| 184 |
+
"loss": 0.6341,
|
| 185 |
+
"rewards/accuracies": 0.737500011920929,
|
| 186 |
+
"rewards/chosen": 0.33271390199661255,
|
| 187 |
+
"rewards/margins": 0.14281830191612244,
|
| 188 |
+
"rewards/rejected": 0.18989557027816772,
|
| 189 |
+
"step": 55
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"epoch": 0.16,
|
| 193 |
+
"grad_norm": 19.192901766300064,
|
| 194 |
+
"learning_rate": 4.948824853131236e-07,
|
| 195 |
+
"logits/chosen": -2.8852925300598145,
|
| 196 |
+
"logits/rejected": -2.859867572784424,
|
| 197 |
+
"logps/chosen": -253.8678741455078,
|
| 198 |
+
"logps/rejected": -212.1206512451172,
|
| 199 |
+
"loss": 0.6257,
|
| 200 |
+
"rewards/accuracies": 0.8125,
|
| 201 |
+
"rewards/chosen": 0.3522695302963257,
|
| 202 |
+
"rewards/margins": 0.1734958440065384,
|
| 203 |
+
"rewards/rejected": 0.1787737011909485,
|
| 204 |
"step": 60
|
| 205 |
},
|
| 206 |
+
{
|
| 207 |
+
"epoch": 0.17,
|
| 208 |
+
"grad_norm": 19.6177452377838,
|
| 209 |
+
"learning_rate": 4.923053649761152e-07,
|
| 210 |
+
"logits/chosen": -2.8169569969177246,
|
| 211 |
+
"logits/rejected": -2.7849762439727783,
|
| 212 |
+
"logps/chosen": -244.50808715820312,
|
| 213 |
+
"logps/rejected": -201.7557830810547,
|
| 214 |
+
"loss": 0.6149,
|
| 215 |
+
"rewards/accuracies": 0.7875000238418579,
|
| 216 |
+
"rewards/chosen": 0.35878387093544006,
|
| 217 |
+
"rewards/margins": 0.19476714730262756,
|
| 218 |
+
"rewards/rejected": 0.1640167087316513,
|
| 219 |
+
"step": 65
|
| 220 |
+
},
|
| 221 |
{
|
| 222 |
"epoch": 0.18,
|
| 223 |
+
"grad_norm": 20.424079652700506,
|
| 224 |
+
"learning_rate": 4.892141805936084e-07,
|
| 225 |
+
"logits/chosen": -2.789267063140869,
|
| 226 |
+
"logits/rejected": -2.73822283744812,
|
| 227 |
+
"logps/chosen": -250.8653564453125,
|
| 228 |
+
"logps/rejected": -224.6582794189453,
|
| 229 |
+
"loss": 0.6149,
|
| 230 |
+
"rewards/accuracies": 0.824999988079071,
|
| 231 |
+
"rewards/chosen": 0.4033736288547516,
|
| 232 |
+
"rewards/margins": 0.23386640846729279,
|
| 233 |
+
"rewards/rejected": 0.1695072501897812,
|
| 234 |
"step": 70
|
| 235 |
},
|
| 236 |
{
|
| 237 |
"epoch": 0.2,
|
| 238 |
+
"grad_norm": 25.663818502097516,
|
| 239 |
+
"learning_rate": 4.856154902818431e-07,
|
| 240 |
+
"logits/chosen": -2.8126118183135986,
|
| 241 |
+
"logits/rejected": -2.7708253860473633,
|
| 242 |
+
"logps/chosen": -229.9053955078125,
|
| 243 |
+
"logps/rejected": -206.87258911132812,
|
| 244 |
+
"loss": 0.6004,
|
| 245 |
+
"rewards/accuracies": 0.75,
|
| 246 |
+
"rewards/chosen": 0.2887989580631256,
|
| 247 |
+
"rewards/margins": 0.22907009720802307,
|
| 248 |
+
"rewards/rejected": 0.05972885340452194,
|
| 249 |
+
"step": 75
|
| 250 |
+
},
|
| 251 |
+
{
|
| 252 |
+
"epoch": 0.21,
|
| 253 |
+
"grad_norm": 24.00962454367728,
|
| 254 |
+
"learning_rate": 4.81516928858564e-07,
|
| 255 |
+
"logits/chosen": -2.7702012062072754,
|
| 256 |
+
"logits/rejected": -2.7202858924865723,
|
| 257 |
+
"logps/chosen": -271.06317138671875,
|
| 258 |
+
"logps/rejected": -220.15475463867188,
|
| 259 |
+
"loss": 0.5964,
|
| 260 |
+
"rewards/accuracies": 0.8125,
|
| 261 |
+
"rewards/chosen": 0.3578353524208069,
|
| 262 |
+
"rewards/margins": 0.31964898109436035,
|
| 263 |
+
"rewards/rejected": 0.03818630054593086,
|
| 264 |
"step": 80
|
| 265 |
},
|
| 266 |
{
|
| 267 |
+
"epoch": 0.22,
|
| 268 |
+
"grad_norm": 26.200013285180134,
|
| 269 |
+
"learning_rate": 4.769271916453385e-07,
|
| 270 |
+
"logits/chosen": -2.817336320877075,
|
| 271 |
+
"logits/rejected": -2.7950100898742676,
|
| 272 |
+
"logps/chosen": -257.9813537597656,
|
| 273 |
+
"logps/rejected": -234.77108764648438,
|
| 274 |
+
"loss": 0.5692,
|
| 275 |
+
"rewards/accuracies": 0.800000011920929,
|
| 276 |
+
"rewards/chosen": 0.16794440150260925,
|
| 277 |
+
"rewards/margins": 0.3747493624687195,
|
| 278 |
+
"rewards/rejected": -0.20680496096611023,
|
| 279 |
+
"step": 85
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"epoch": 0.24,
|
| 283 |
+
"grad_norm": 31.71282371862922,
|
| 284 |
+
"learning_rate": 4.7185601601995784e-07,
|
| 285 |
+
"logits/chosen": -2.8321759700775146,
|
| 286 |
+
"logits/rejected": -2.8417580127716064,
|
| 287 |
+
"logps/chosen": -252.80313110351562,
|
| 288 |
+
"logps/rejected": -233.3778839111328,
|
| 289 |
+
"loss": 0.5452,
|
| 290 |
+
"rewards/accuracies": 0.8999999761581421,
|
| 291 |
+
"rewards/chosen": 0.17818805575370789,
|
| 292 |
+
"rewards/margins": 0.4816998839378357,
|
| 293 |
+
"rewards/rejected": -0.3035118579864502,
|
| 294 |
"step": 90
|
| 295 |
},
|
| 296 |
+
{
|
| 297 |
+
"epoch": 0.25,
|
| 298 |
+
"grad_norm": 30.203747907953968,
|
| 299 |
+
"learning_rate": 4.6631416075805886e-07,
|
| 300 |
+
"logits/chosen": -2.9183051586151123,
|
| 301 |
+
"logits/rejected": -2.8942832946777344,
|
| 302 |
+
"logps/chosen": -292.6710510253906,
|
| 303 |
+
"logps/rejected": -278.3623962402344,
|
| 304 |
+
"loss": 0.5293,
|
| 305 |
+
"rewards/accuracies": 0.762499988079071,
|
| 306 |
+
"rewards/chosen": 0.0014784678351134062,
|
| 307 |
+
"rewards/margins": 0.45240503549575806,
|
| 308 |
+
"rewards/rejected": -0.4509265422821045,
|
| 309 |
+
"step": 95
|
| 310 |
+
},
|
| 311 |
{
|
| 312 |
"epoch": 0.26,
|
| 313 |
+
"grad_norm": 35.20356531841964,
|
| 314 |
+
"learning_rate": 4.603133832077953e-07,
|
| 315 |
+
"logits/chosen": -3.0372838973999023,
|
| 316 |
+
"logits/rejected": -3.03391432762146,
|
| 317 |
+
"logps/chosen": -264.54412841796875,
|
| 318 |
+
"logps/rejected": -271.9236755371094,
|
| 319 |
+
"loss": 0.5292,
|
| 320 |
+
"rewards/accuracies": 0.7749999761581421,
|
| 321 |
+
"rewards/chosen": -0.24657364189624786,
|
| 322 |
+
"rewards/margins": 0.4146839678287506,
|
| 323 |
+
"rewards/rejected": -0.6612575650215149,
|
| 324 |
"step": 100
|
| 325 |
},
|
| 326 |
{
|
| 327 |
"epoch": 0.26,
|
| 328 |
+
"eval_logits/chosen": -2.981271743774414,
|
| 329 |
+
"eval_logits/rejected": -3.007460117340088,
|
| 330 |
+
"eval_logps/chosen": -500.8688049316406,
|
| 331 |
+
"eval_logps/rejected": -504.0709228515625,
|
| 332 |
+
"eval_loss": 0.6958277821540833,
|
| 333 |
+
"eval_rewards/accuracies": 0.53125,
|
| 334 |
+
"eval_rewards/chosen": -1.2149137258529663,
|
| 335 |
+
"eval_rewards/margins": 0.09968078136444092,
|
| 336 |
+
"eval_rewards/rejected": -1.3145945072174072,
|
| 337 |
+
"eval_runtime": 97.6692,
|
| 338 |
+
"eval_samples_per_second": 20.477,
|
| 339 |
+
"eval_steps_per_second": 0.328,
|
| 340 |
"step": 100
|
| 341 |
},
|
| 342 |
{
|
| 343 |
"epoch": 0.28,
|
| 344 |
+
"grad_norm": 35.37658079116714,
|
| 345 |
+
"learning_rate": 4.538664143459818e-07,
|
| 346 |
+
"logits/chosen": -3.1525657176971436,
|
| 347 |
+
"logits/rejected": -3.170300245285034,
|
| 348 |
+
"logps/chosen": -301.29736328125,
|
| 349 |
+
"logps/rejected": -306.07171630859375,
|
| 350 |
+
"loss": 0.4875,
|
| 351 |
+
"rewards/accuracies": 0.8374999761581421,
|
| 352 |
+
"rewards/chosen": -0.40715986490249634,
|
| 353 |
+
"rewards/margins": 0.6011335849761963,
|
| 354 |
+
"rewards/rejected": -1.0082933902740479,
|
| 355 |
+
"step": 105
|
| 356 |
+
},
|
| 357 |
+
{
|
| 358 |
+
"epoch": 0.29,
|
| 359 |
+
"grad_norm": 44.5542186040465,
|
| 360 |
+
"learning_rate": 4.4698693176863316e-07,
|
| 361 |
+
"logits/chosen": -3.1490285396575928,
|
| 362 |
+
"logits/rejected": -3.167726993560791,
|
| 363 |
+
"logps/chosen": -326.18792724609375,
|
| 364 |
+
"logps/rejected": -354.35784912109375,
|
| 365 |
+
"loss": 0.4645,
|
| 366 |
+
"rewards/accuracies": 0.800000011920929,
|
| 367 |
+
"rewards/chosen": -0.6683047413825989,
|
| 368 |
+
"rewards/margins": 0.806756854057312,
|
| 369 |
+
"rewards/rejected": -1.4750616550445557,
|
| 370 |
"step": 110
|
| 371 |
},
|
| 372 |
{
|
| 373 |
+
"epoch": 0.3,
|
| 374 |
+
"grad_norm": 46.11054779372386,
|
| 375 |
+
"learning_rate": 4.396895306731977e-07,
|
| 376 |
+
"logits/chosen": -3.2922375202178955,
|
| 377 |
+
"logits/rejected": -3.3002562522888184,
|
| 378 |
+
"logps/chosen": -316.46331787109375,
|
| 379 |
+
"logps/rejected": -344.7227783203125,
|
| 380 |
+
"loss": 0.478,
|
| 381 |
"rewards/accuracies": 0.824999988079071,
|
| 382 |
+
"rewards/chosen": -0.5892685651779175,
|
| 383 |
+
"rewards/margins": 0.8242443203926086,
|
| 384 |
+
"rewards/rejected": -1.413512945175171,
|
| 385 |
+
"step": 115
|
| 386 |
+
},
|
| 387 |
+
{
|
| 388 |
+
"epoch": 0.32,
|
| 389 |
+
"grad_norm": 39.587491646799435,
|
| 390 |
+
"learning_rate": 4.319896928940505e-07,
|
| 391 |
+
"logits/chosen": -3.4349207878112793,
|
| 392 |
+
"logits/rejected": -3.4556515216827393,
|
| 393 |
+
"logps/chosen": -430.2960510253906,
|
| 394 |
+
"logps/rejected": -467.32061767578125,
|
| 395 |
+
"loss": 0.4502,
|
| 396 |
+
"rewards/accuracies": 0.7749999761581421,
|
| 397 |
+
"rewards/chosen": -1.621132254600525,
|
| 398 |
+
"rewards/margins": 0.7838995456695557,
|
| 399 |
+
"rewards/rejected": -2.405031681060791,
|
| 400 |
"step": 120
|
| 401 |
},
|
| 402 |
{
|
| 403 |
"epoch": 0.33,
|
| 404 |
+
"grad_norm": 36.84462226750656,
|
| 405 |
+
"learning_rate": 4.2390375405693723e-07,
|
| 406 |
+
"logits/chosen": -3.474997043609619,
|
| 407 |
+
"logits/rejected": -3.5272536277770996,
|
| 408 |
+
"logps/chosen": -415.80438232421875,
|
| 409 |
+
"logps/rejected": -468.0596618652344,
|
| 410 |
+
"loss": 0.4422,
|
| 411 |
+
"rewards/accuracies": 0.762499988079071,
|
| 412 |
+
"rewards/chosen": -1.5106348991394043,
|
| 413 |
+
"rewards/margins": 1.0294691324234009,
|
| 414 |
+
"rewards/rejected": -2.5401039123535156,
|
| 415 |
+
"step": 125
|
| 416 |
+
},
|
| 417 |
+
{
|
| 418 |
+
"epoch": 0.34,
|
| 419 |
+
"grad_norm": 47.09166107339635,
|
| 420 |
+
"learning_rate": 4.1544886892205354e-07,
|
| 421 |
+
"logits/chosen": -3.5425872802734375,
|
| 422 |
+
"logits/rejected": -3.5473670959472656,
|
| 423 |
+
"logps/chosen": -377.5634765625,
|
| 424 |
+
"logps/rejected": -425.98016357421875,
|
| 425 |
+
"loss": 0.4589,
|
| 426 |
"rewards/accuracies": 0.8125,
|
| 427 |
+
"rewards/chosen": -0.8434383273124695,
|
| 428 |
+
"rewards/margins": 0.9450035095214844,
|
| 429 |
+
"rewards/rejected": -1.7884416580200195,
|
| 430 |
"step": 130
|
| 431 |
},
|
| 432 |
{
|
| 433 |
"epoch": 0.36,
|
| 434 |
+
"grad_norm": 49.77750455648269,
|
| 435 |
+
"learning_rate": 4.0664297498928534e-07,
|
| 436 |
+
"logits/chosen": -3.6749653816223145,
|
| 437 |
+
"logits/rejected": -3.6941752433776855,
|
| 438 |
+
"logps/chosen": -392.3956604003906,
|
| 439 |
+
"logps/rejected": -454.9180603027344,
|
| 440 |
+
"loss": 0.4261,
|
| 441 |
+
"rewards/accuracies": 0.800000011920929,
|
| 442 |
+
"rewards/chosen": -1.5022857189178467,
|
| 443 |
+
"rewards/margins": 1.0955121517181396,
|
| 444 |
+
"rewards/rejected": -2.5977978706359863,
|
| 445 |
+
"step": 135
|
| 446 |
+
},
|
| 447 |
+
{
|
| 448 |
+
"epoch": 0.37,
|
| 449 |
+
"grad_norm": 49.72958773280421,
|
| 450 |
+
"learning_rate": 3.975047544428254e-07,
|
| 451 |
+
"logits/chosen": -3.7381629943847656,
|
| 452 |
+
"logits/rejected": -3.768315076828003,
|
| 453 |
+
"logps/chosen": -459.23455810546875,
|
| 454 |
+
"logps/rejected": -531.2666625976562,
|
| 455 |
+
"loss": 0.4262,
|
| 456 |
+
"rewards/accuracies": 0.8125,
|
| 457 |
+
"rewards/chosen": -1.8736200332641602,
|
| 458 |
+
"rewards/margins": 1.138641119003296,
|
| 459 |
+
"rewards/rejected": -3.012261152267456,
|
| 460 |
"step": 140
|
| 461 |
},
|
| 462 |
{
|
| 463 |
"epoch": 0.38,
|
| 464 |
+
"grad_norm": 37.39808184046548,
|
| 465 |
+
"learning_rate": 3.880535945158997e-07,
|
| 466 |
+
"logits/chosen": -3.7747676372528076,
|
| 467 |
+
"logits/rejected": -3.7997519969940186,
|
| 468 |
+
"logps/chosen": -367.59576416015625,
|
| 469 |
+
"logps/rejected": -449.857177734375,
|
| 470 |
+
"loss": 0.4127,
|
| 471 |
+
"rewards/accuracies": 0.7749999761581421,
|
| 472 |
+
"rewards/chosen": -1.2651011943817139,
|
| 473 |
+
"rewards/margins": 1.1386528015136719,
|
| 474 |
+
"rewards/rejected": -2.403754234313965,
|
| 475 |
+
"step": 145
|
| 476 |
+
},
|
| 477 |
+
{
|
| 478 |
+
"epoch": 0.4,
|
| 479 |
+
"grad_norm": 47.294132664555036,
|
| 480 |
+
"learning_rate": 3.78309546359696e-07,
|
| 481 |
+
"logits/chosen": -3.9040164947509766,
|
| 482 |
+
"logits/rejected": -3.91438627243042,
|
| 483 |
+
"logps/chosen": -410.07159423828125,
|
| 484 |
+
"logps/rejected": -498.1920471191406,
|
| 485 |
+
"loss": 0.4235,
|
| 486 |
+
"rewards/accuracies": 0.7875000238418579,
|
| 487 |
+
"rewards/chosen": -1.4813024997711182,
|
| 488 |
+
"rewards/margins": 1.1341646909713745,
|
| 489 |
+
"rewards/rejected": -2.6154673099517822,
|
| 490 |
"step": 150
|
| 491 |
},
|
| 492 |
{
|
| 493 |
"epoch": 0.41,
|
| 494 |
+
"grad_norm": 40.77943292899045,
|
| 495 |
+
"learning_rate": 3.6829328250375227e-07,
|
| 496 |
+
"logits/chosen": -3.932398557662964,
|
| 497 |
+
"logits/rejected": -4.020986080169678,
|
| 498 |
+
"logps/chosen": -427.6490783691406,
|
| 499 |
+
"logps/rejected": -508.22930908203125,
|
| 500 |
+
"loss": 0.3809,
|
| 501 |
+
"rewards/accuracies": 0.824999988079071,
|
| 502 |
+
"rewards/chosen": -1.581094741821289,
|
| 503 |
+
"rewards/margins": 1.3298404216766357,
|
| 504 |
+
"rewards/rejected": -2.910935401916504,
|
| 505 |
+
"step": 155
|
| 506 |
+
},
|
| 507 |
+
{
|
| 508 |
+
"epoch": 0.42,
|
| 509 |
+
"grad_norm": 41.2026716936047,
|
| 510 |
+
"learning_rate": 3.580260529980584e-07,
|
| 511 |
+
"logits/chosen": -4.01393461227417,
|
| 512 |
+
"logits/rejected": -4.080017566680908,
|
| 513 |
+
"logps/chosen": -387.573974609375,
|
| 514 |
+
"logps/rejected": -487.1441345214844,
|
| 515 |
+
"loss": 0.3859,
|
| 516 |
+
"rewards/accuracies": 0.7749999761581421,
|
| 517 |
+
"rewards/chosen": -1.2360130548477173,
|
| 518 |
+
"rewards/margins": 1.409085988998413,
|
| 519 |
+
"rewards/rejected": -2.64509916305542,
|
| 520 |
"step": 160
|
| 521 |
},
|
| 522 |
{
|
| 523 |
"epoch": 0.43,
|
| 524 |
+
"grad_norm": 40.85253910181408,
|
| 525 |
+
"learning_rate": 3.475296403299163e-07,
|
| 526 |
+
"logits/chosen": -4.092155456542969,
|
| 527 |
+
"logits/rejected": -4.100491523742676,
|
| 528 |
+
"logps/chosen": -388.4344482421875,
|
| 529 |
+
"logps/rejected": -494.84771728515625,
|
| 530 |
+
"loss": 0.391,
|
| 531 |
+
"rewards/accuracies": 0.824999988079071,
|
| 532 |
+
"rewards/chosen": -1.4383682012557983,
|
| 533 |
+
"rewards/margins": 1.5076179504394531,
|
| 534 |
+
"rewards/rejected": -2.945986032485962,
|
| 535 |
+
"step": 165
|
| 536 |
+
},
|
| 537 |
+
{
|
| 538 |
+
"epoch": 0.45,
|
| 539 |
+
"grad_norm": 43.630879623871635,
|
| 540 |
+
"learning_rate": 3.36826313211205e-07,
|
| 541 |
+
"logits/chosen": -4.23541784286499,
|
| 542 |
+
"logits/rejected": -4.378731727600098,
|
| 543 |
+
"logps/chosen": -423.3770446777344,
|
| 544 |
+
"logps/rejected": -521.7530517578125,
|
| 545 |
+
"loss": 0.3784,
|
| 546 |
+
"rewards/accuracies": 0.7749999761581421,
|
| 547 |
+
"rewards/chosen": -1.7406959533691406,
|
| 548 |
+
"rewards/margins": 1.5629594326019287,
|
| 549 |
+
"rewards/rejected": -3.3036551475524902,
|
| 550 |
"step": 170
|
| 551 |
},
|
| 552 |
{
|
| 553 |
"epoch": 0.46,
|
| 554 |
+
"grad_norm": 40.910837008073884,
|
| 555 |
+
"learning_rate": 3.259387793340943e-07,
|
| 556 |
+
"logits/chosen": -4.390842437744141,
|
| 557 |
+
"logits/rejected": -4.466560363769531,
|
| 558 |
+
"logps/chosen": -482.41748046875,
|
| 559 |
+
"logps/rejected": -587.3478393554688,
|
| 560 |
+
"loss": 0.3845,
|
| 561 |
+
"rewards/accuracies": 0.7875000238418579,
|
| 562 |
+
"rewards/chosen": -2.141714096069336,
|
| 563 |
+
"rewards/margins": 1.4235177040100098,
|
| 564 |
+
"rewards/rejected": -3.5652313232421875,
|
| 565 |
+
"step": 175
|
| 566 |
+
},
|
| 567 |
+
{
|
| 568 |
+
"epoch": 0.47,
|
| 569 |
+
"grad_norm": 43.846997204513016,
|
| 570 |
+
"learning_rate": 3.14890137195437e-07,
|
| 571 |
+
"logits/chosen": -4.389448642730713,
|
| 572 |
+
"logits/rejected": -4.424112319946289,
|
| 573 |
+
"logps/chosen": -481.9625549316406,
|
| 574 |
+
"logps/rejected": -567.7194213867188,
|
| 575 |
+
"loss": 0.3682,
|
| 576 |
+
"rewards/accuracies": 0.800000011920929,
|
| 577 |
+
"rewards/chosen": -2.1195263862609863,
|
| 578 |
+
"rewards/margins": 1.2820372581481934,
|
| 579 |
+
"rewards/rejected": -3.4015636444091797,
|
| 580 |
"step": 180
|
| 581 |
},
|
| 582 |
{
|
| 583 |
"epoch": 0.49,
|
| 584 |
+
"grad_norm": 48.89695189111101,
|
| 585 |
+
"learning_rate": 3.0370382709204883e-07,
|
| 586 |
+
"logits/chosen": -4.416136264801025,
|
| 587 |
+
"logits/rejected": -4.519248962402344,
|
| 588 |
+
"logps/chosen": -472.10931396484375,
|
| 589 |
+
"logps/rejected": -609.0546264648438,
|
| 590 |
+
"loss": 0.3879,
|
| 591 |
+
"rewards/accuracies": 0.887499988079071,
|
| 592 |
+
"rewards/chosen": -1.9735286235809326,
|
| 593 |
+
"rewards/margins": 1.7880016565322876,
|
| 594 |
+
"rewards/rejected": -3.7615303993225098,
|
| 595 |
+
"step": 185
|
| 596 |
+
},
|
| 597 |
+
{
|
| 598 |
+
"epoch": 0.5,
|
| 599 |
+
"grad_norm": 49.97288424600573,
|
| 600 |
+
"learning_rate": 2.9240358139084013e-07,
|
| 601 |
+
"logits/chosen": -4.548556804656982,
|
| 602 |
+
"logits/rejected": -4.62185001373291,
|
| 603 |
+
"logps/chosen": -421.85052490234375,
|
| 604 |
+
"logps/rejected": -534.3924560546875,
|
| 605 |
+
"loss": 0.3824,
|
| 606 |
+
"rewards/accuracies": 0.800000011920929,
|
| 607 |
+
"rewards/chosen": -1.734278917312622,
|
| 608 |
+
"rewards/margins": 1.5305709838867188,
|
| 609 |
+
"rewards/rejected": -3.264849901199341,
|
| 610 |
"step": 190
|
| 611 |
},
|
| 612 |
{
|
| 613 |
"epoch": 0.51,
|
| 614 |
+
"grad_norm": 42.794361716855846,
|
| 615 |
+
"learning_rate": 2.810133741793052e-07,
|
| 616 |
+
"logits/chosen": -4.521183490753174,
|
| 617 |
+
"logits/rejected": -4.65042781829834,
|
| 618 |
+
"logps/chosen": -472.19500732421875,
|
| 619 |
+
"logps/rejected": -594.0242309570312,
|
| 620 |
+
"loss": 0.3875,
|
| 621 |
+
"rewards/accuracies": 0.862500011920929,
|
| 622 |
+
"rewards/chosen": -2.0321810245513916,
|
| 623 |
+
"rewards/margins": 1.7259247303009033,
|
| 624 |
+
"rewards/rejected": -3.758105516433716,
|
| 625 |
+
"step": 195
|
| 626 |
+
},
|
| 627 |
+
{
|
| 628 |
+
"epoch": 0.53,
|
| 629 |
+
"grad_norm": 43.28984360773978,
|
| 630 |
+
"learning_rate": 2.695573704031885e-07,
|
| 631 |
+
"logits/chosen": -4.522828102111816,
|
| 632 |
+
"logits/rejected": -4.58953332901001,
|
| 633 |
+
"logps/chosen": -470.6817321777344,
|
| 634 |
+
"logps/rejected": -592.047119140625,
|
| 635 |
+
"loss": 0.3733,
|
| 636 |
+
"rewards/accuracies": 0.8374999761581421,
|
| 637 |
+
"rewards/chosen": -1.8410589694976807,
|
| 638 |
+
"rewards/margins": 1.6147441864013672,
|
| 639 |
+
"rewards/rejected": -3.4558029174804688,
|
| 640 |
"step": 200
|
| 641 |
},
|
| 642 |
{
|
| 643 |
+
"epoch": 0.53,
|
| 644 |
+
"eval_logits/chosen": -4.407002925872803,
|
| 645 |
+
"eval_logits/rejected": -4.5463128089904785,
|
| 646 |
+
"eval_logps/chosen": -804.1498413085938,
|
| 647 |
+
"eval_logps/rejected": -826.000732421875,
|
| 648 |
+
"eval_loss": 0.9587702751159668,
|
| 649 |
+
"eval_rewards/accuracies": 0.52734375,
|
| 650 |
+
"eval_rewards/chosen": -4.247724533081055,
|
| 651 |
+
"eval_rewards/margins": 0.28616809844970703,
|
| 652 |
+
"eval_rewards/rejected": -4.53389310836792,
|
| 653 |
+
"eval_runtime": 97.528,
|
| 654 |
+
"eval_samples_per_second": 20.507,
|
| 655 |
+
"eval_steps_per_second": 0.328,
|
| 656 |
"step": 200
|
| 657 |
},
|
| 658 |
{
|
| 659 |
"epoch": 0.54,
|
| 660 |
+
"grad_norm": 45.90321806356986,
|
| 661 |
+
"learning_rate": 2.580598745992342e-07,
|
| 662 |
+
"logits/chosen": -4.592051029205322,
|
| 663 |
+
"logits/rejected": -4.729592800140381,
|
| 664 |
+
"logps/chosen": -462.078857421875,
|
| 665 |
+
"logps/rejected": -578.6028442382812,
|
| 666 |
+
"loss": 0.3683,
|
| 667 |
+
"rewards/accuracies": 0.8500000238418579,
|
| 668 |
+
"rewards/chosen": -1.6925300359725952,
|
| 669 |
+
"rewards/margins": 1.7680349349975586,
|
| 670 |
+
"rewards/rejected": -3.4605648517608643,
|
| 671 |
+
"step": 205
|
| 672 |
+
},
|
| 673 |
+
{
|
| 674 |
+
"epoch": 0.55,
|
| 675 |
+
"grad_norm": 45.69561704557737,
|
| 676 |
+
"learning_rate": 2.465452793317865e-07,
|
| 677 |
+
"logits/chosen": -4.680180549621582,
|
| 678 |
+
"logits/rejected": -4.785167694091797,
|
| 679 |
+
"logps/chosen": -477.2386779785156,
|
| 680 |
+
"logps/rejected": -609.8983154296875,
|
| 681 |
+
"loss": 0.3534,
|
| 682 |
+
"rewards/accuracies": 0.7749999761581421,
|
| 683 |
+
"rewards/chosen": -1.8637679815292358,
|
| 684 |
+
"rewards/margins": 1.7234246730804443,
|
| 685 |
+
"rewards/rejected": -3.587193012237549,
|
| 686 |
"step": 210
|
| 687 |
},
|
| 688 |
{
|
| 689 |
+
"epoch": 0.57,
|
| 690 |
+
"grad_norm": 50.566593421325194,
|
| 691 |
+
"learning_rate": 2.3503801344263344e-07,
|
| 692 |
+
"logits/chosen": -4.860222339630127,
|
| 693 |
+
"logits/rejected": -4.944591045379639,
|
| 694 |
+
"logps/chosen": -456.22357177734375,
|
| 695 |
+
"logps/rejected": -573.6387939453125,
|
| 696 |
+
"loss": 0.3712,
|
| 697 |
+
"rewards/accuracies": 0.862500011920929,
|
| 698 |
+
"rewards/chosen": -2.1773338317871094,
|
| 699 |
+
"rewards/margins": 1.6265901327133179,
|
| 700 |
+
"rewards/rejected": -3.8039238452911377,
|
| 701 |
+
"step": 215
|
| 702 |
+
},
|
| 703 |
+
{
|
| 704 |
+
"epoch": 0.58,
|
| 705 |
+
"grad_norm": 49.14997910435826,
|
| 706 |
+
"learning_rate": 2.2356249022388789e-07,
|
| 707 |
+
"logits/chosen": -4.710982322692871,
|
| 708 |
+
"logits/rejected": -4.878017425537109,
|
| 709 |
+
"logps/chosen": -468.77978515625,
|
| 710 |
+
"logps/rejected": -580.1436157226562,
|
| 711 |
+
"loss": 0.3706,
|
| 712 |
+
"rewards/accuracies": 0.8125,
|
| 713 |
+
"rewards/chosen": -1.8084490299224854,
|
| 714 |
+
"rewards/margins": 1.884009599685669,
|
| 715 |
+
"rewards/rejected": -3.6924586296081543,
|
| 716 |
"step": 220
|
| 717 |
},
|
| 718 |
{
|
| 719 |
"epoch": 0.59,
|
| 720 |
+
"grad_norm": 48.71640334228914,
|
| 721 |
+
"learning_rate": 2.121430556238559e-07,
|
| 722 |
+
"logits/chosen": -4.8175554275512695,
|
| 723 |
+
"logits/rejected": -4.996689319610596,
|
| 724 |
+
"logps/chosen": -471.17132568359375,
|
| 725 |
+
"logps/rejected": -604.3876342773438,
|
| 726 |
+
"loss": 0.3386,
|
| 727 |
"rewards/accuracies": 0.862500011920929,
|
| 728 |
+
"rewards/chosen": -2.008169412612915,
|
| 729 |
+
"rewards/margins": 1.8934684991836548,
|
| 730 |
+
"rewards/rejected": -3.9016380310058594,
|
| 731 |
+
"step": 225
|
| 732 |
},
|
| 733 |
{
|
| 734 |
"epoch": 0.61,
|
| 735 |
+
"grad_norm": 39.742825593785554,
|
| 736 |
+
"learning_rate": 2.0080393659578038e-07,
|
| 737 |
+
"logits/chosen": -4.740202903747559,
|
| 738 |
+
"logits/rejected": -5.004001617431641,
|
| 739 |
+
"logps/chosen": -513.0294189453125,
|
| 740 |
+
"logps/rejected": -656.8961181640625,
|
| 741 |
+
"loss": 0.3492,
|
| 742 |
+
"rewards/accuracies": 0.800000011920929,
|
| 743 |
+
"rewards/chosen": -2.400604248046875,
|
| 744 |
+
"rewards/margins": 2.0355873107910156,
|
| 745 |
+
"rewards/rejected": -4.436191558837891,
|
| 746 |
+
"step": 230
|
| 747 |
+
},
|
| 748 |
+
{
|
| 749 |
+
"epoch": 0.62,
|
| 750 |
+
"grad_norm": 50.20217799621439,
|
| 751 |
+
"learning_rate": 1.895691896990388e-07,
|
| 752 |
+
"logits/chosen": -4.822530269622803,
|
| 753 |
+
"logits/rejected": -4.9560699462890625,
|
| 754 |
+
"logps/chosen": -474.0023498535156,
|
| 755 |
+
"logps/rejected": -579.5379638671875,
|
| 756 |
+
"loss": 0.3354,
|
| 757 |
+
"rewards/accuracies": 0.800000011920929,
|
| 758 |
+
"rewards/chosen": -2.1355338096618652,
|
| 759 |
+
"rewards/margins": 1.5200008153915405,
|
| 760 |
+
"rewards/rejected": -3.655534267425537,
|
| 761 |
+
"step": 235
|
| 762 |
+
},
|
| 763 |
+
{
|
| 764 |
+
"epoch": 0.63,
|
| 765 |
+
"grad_norm": 43.0245343621866,
|
| 766 |
+
"learning_rate": 1.7846265006183976e-07,
|
| 767 |
+
"logits/chosen": -4.6783013343811035,
|
| 768 |
+
"logits/rejected": -4.8442277908325195,
|
| 769 |
+
"logps/chosen": -524.712158203125,
|
| 770 |
+
"logps/rejected": -632.6328125,
|
| 771 |
+
"loss": 0.3455,
|
| 772 |
+
"rewards/accuracies": 0.862500011920929,
|
| 773 |
+
"rewards/chosen": -2.490675687789917,
|
| 774 |
+
"rewards/margins": 1.6402429342269897,
|
| 775 |
+
"rewards/rejected": -4.130918979644775,
|
| 776 |
"step": 240
|
| 777 |
},
|
| 778 |
{
|
| 779 |
+
"epoch": 0.65,
|
| 780 |
+
"grad_norm": 50.41005236551354,
|
| 781 |
+
"learning_rate": 1.6750788081369948e-07,
|
| 782 |
+
"logits/chosen": -4.760382175445557,
|
| 783 |
+
"logits/rejected": -4.9731292724609375,
|
| 784 |
+
"logps/chosen": -475.59765625,
|
| 785 |
+
"logps/rejected": -638.8780517578125,
|
| 786 |
+
"loss": 0.3533,
|
| 787 |
+
"rewards/accuracies": 0.887499988079071,
|
| 788 |
+
"rewards/chosen": -2.0123469829559326,
|
| 789 |
+
"rewards/margins": 2.291203737258911,
|
| 790 |
+
"rewards/rejected": -4.303550720214844,
|
| 791 |
+
"step": 245
|
| 792 |
},
|
| 793 |
{
|
| 794 |
"epoch": 0.66,
|
| 795 |
+
"grad_norm": 41.2044329360216,
|
| 796 |
+
"learning_rate": 1.5672812309497722e-07,
|
| 797 |
+
"logits/chosen": -4.802388668060303,
|
| 798 |
+
"logits/rejected": -4.928206443786621,
|
| 799 |
+
"logps/chosen": -410.943115234375,
|
| 800 |
+
"logps/rejected": -523.5797119140625,
|
| 801 |
+
"loss": 0.3523,
|
| 802 |
+
"rewards/accuracies": 0.8374999761581421,
|
| 803 |
+
"rewards/chosen": -1.6920316219329834,
|
| 804 |
+
"rewards/margins": 1.4624649286270142,
|
| 805 |
+
"rewards/rejected": -3.154496669769287,
|
| 806 |
+
"step": 250
|
| 807 |
+
},
|
| 808 |
+
{
|
| 809 |
+
"epoch": 0.67,
|
| 810 |
+
"grad_norm": 44.69671736661672,
|
| 811 |
+
"learning_rate": 1.461462467495284e-07,
|
| 812 |
+
"logits/chosen": -4.758819580078125,
|
| 813 |
+
"logits/rejected": -4.947306156158447,
|
| 814 |
+
"logps/chosen": -495.4164123535156,
|
| 815 |
+
"logps/rejected": -593.2689208984375,
|
| 816 |
+
"loss": 0.3703,
|
| 817 |
+
"rewards/accuracies": 0.7875000238418579,
|
| 818 |
+
"rewards/chosen": -2.218562602996826,
|
| 819 |
+
"rewards/margins": 1.439822793006897,
|
| 820 |
+
"rewards/rejected": -3.6583850383758545,
|
| 821 |
+
"step": 255
|
| 822 |
},
|
| 823 |
{
|
| 824 |
"epoch": 0.69,
|
| 825 |
+
"grad_norm": 47.815462029236265,
|
| 826 |
+
"learning_rate": 1.357847018050843e-07,
|
| 827 |
+
"logits/chosen": -4.747325420379639,
|
| 828 |
+
"logits/rejected": -4.958649158477783,
|
| 829 |
+
"logps/chosen": -592.2781372070312,
|
| 830 |
+
"logps/rejected": -758.1558837890625,
|
| 831 |
+
"loss": 0.3513,
|
| 832 |
+
"rewards/accuracies": 0.875,
|
| 833 |
+
"rewards/chosen": -2.605052947998047,
|
| 834 |
+
"rewards/margins": 2.2088265419006348,
|
| 835 |
+
"rewards/rejected": -4.81387996673584,
|
| 836 |
+
"step": 260
|
| 837 |
+
},
|
| 838 |
+
{
|
| 839 |
+
"epoch": 0.7,
|
| 840 |
+
"grad_norm": 49.07485622430381,
|
| 841 |
+
"learning_rate": 1.2566547084429324e-07,
|
| 842 |
+
"logits/chosen": -4.789057731628418,
|
| 843 |
+
"logits/rejected": -5.023972988128662,
|
| 844 |
+
"logps/chosen": -472.7372131347656,
|
| 845 |
+
"logps/rejected": -615.0753784179688,
|
| 846 |
+
"loss": 0.3591,
|
| 847 |
+
"rewards/accuracies": 0.800000011920929,
|
| 848 |
+
"rewards/chosen": -2.2092373371124268,
|
| 849 |
+
"rewards/margins": 1.8947007656097412,
|
| 850 |
+
"rewards/rejected": -4.103938579559326,
|
| 851 |
+
"step": 265
|
| 852 |
+
},
|
| 853 |
+
{
|
| 854 |
+
"epoch": 0.71,
|
| 855 |
+
"grad_norm": 45.44279309151153,
|
| 856 |
+
"learning_rate": 1.1581002236747328e-07,
|
| 857 |
+
"logits/chosen": -4.7014241218566895,
|
| 858 |
+
"logits/rejected": -4.898508548736572,
|
| 859 |
+
"logps/chosen": -470.0,
|
| 860 |
+
"logps/rejected": -624.503662109375,
|
| 861 |
+
"loss": 0.368,
|
| 862 |
+
"rewards/accuracies": 0.887499988079071,
|
| 863 |
+
"rewards/chosen": -2.254678249359131,
|
| 864 |
+
"rewards/margins": 2.110576868057251,
|
| 865 |
+
"rewards/rejected": -4.365254878997803,
|
| 866 |
"step": 270
|
| 867 |
},
|
| 868 |
{
|
| 869 |
"epoch": 0.72,
|
| 870 |
+
"grad_norm": 53.087469451573526,
|
| 871 |
+
"learning_rate": 1.062392652460177e-07,
|
| 872 |
+
"logits/chosen": -4.696314811706543,
|
| 873 |
+
"logits/rejected": -4.83748197555542,
|
| 874 |
+
"logps/chosen": -505.9961853027344,
|
| 875 |
+
"logps/rejected": -626.0050048828125,
|
| 876 |
+
"loss": 0.3245,
|
| 877 |
+
"rewards/accuracies": 0.862500011920929,
|
| 878 |
+
"rewards/chosen": -1.9974615573883057,
|
| 879 |
+
"rewards/margins": 1.828150987625122,
|
| 880 |
+
"rewards/rejected": -3.8256123065948486,
|
| 881 |
+
"step": 275
|
| 882 |
},
|
| 883 |
{
|
| 884 |
"epoch": 0.74,
|
| 885 |
+
"grad_norm": 41.5564187525106,
|
| 886 |
+
"learning_rate": 9.697350436308427e-08,
|
| 887 |
+
"logits/chosen": -4.737555027008057,
|
| 888 |
+
"logits/rejected": -4.948160648345947,
|
| 889 |
+
"logps/chosen": -507.66607666015625,
|
| 890 |
+
"logps/rejected": -594.7119750976562,
|
| 891 |
+
"loss": 0.3584,
|
| 892 |
+
"rewards/accuracies": 0.8125,
|
| 893 |
+
"rewards/chosen": -2.3149771690368652,
|
| 894 |
+
"rewards/margins": 1.4441829919815063,
|
| 895 |
+
"rewards/rejected": -3.759159803390503,
|
| 896 |
+
"step": 280
|
| 897 |
+
},
|
| 898 |
+
{
|
| 899 |
+
"epoch": 0.75,
|
| 900 |
+
"grad_norm": 44.45010163032659,
|
| 901 |
+
"learning_rate": 8.803239753567829e-08,
|
| 902 |
+
"logits/chosen": -4.868664741516113,
|
| 903 |
+
"logits/rejected": -4.975742340087891,
|
| 904 |
+
"logps/chosen": -469.70263671875,
|
| 905 |
+
"logps/rejected": -589.5235595703125,
|
| 906 |
+
"loss": 0.3498,
|
| 907 |
+
"rewards/accuracies": 0.875,
|
| 908 |
+
"rewards/chosen": -2.1795949935913086,
|
| 909 |
+
"rewards/margins": 1.7358201742172241,
|
| 910 |
+
"rewards/rejected": -3.9154155254364014,
|
| 911 |
+
"step": 285
|
| 912 |
+
},
|
| 913 |
+
{
|
| 914 |
+
"epoch": 0.76,
|
| 915 |
+
"grad_norm": 38.027371906598745,
|
| 916 |
+
"learning_rate": 7.943491380952188e-08,
|
| 917 |
+
"logits/chosen": -4.954745292663574,
|
| 918 |
+
"logits/rejected": -5.065755844116211,
|
| 919 |
+
"logps/chosen": -429.4737243652344,
|
| 920 |
+
"logps/rejected": -571.6141967773438,
|
| 921 |
+
"loss": 0.3285,
|
| 922 |
+
"rewards/accuracies": 0.862500011920929,
|
| 923 |
+
"rewards/chosen": -1.8640539646148682,
|
| 924 |
+
"rewards/margins": 1.8129631280899048,
|
| 925 |
+
"rewards/rejected": -3.6770172119140625,
|
| 926 |
"step": 290
|
| 927 |
},
|
| 928 |
{
|
| 929 |
+
"epoch": 0.78,
|
| 930 |
+
"grad_norm": 47.705363932547336,
|
| 931 |
+
"learning_rate": 7.119929321518875e-08,
|
| 932 |
+
"logits/chosen": -4.735751152038574,
|
| 933 |
+
"logits/rejected": -5.044283866882324,
|
| 934 |
+
"logps/chosen": -445.70367431640625,
|
| 935 |
+
"logps/rejected": -583.5177612304688,
|
| 936 |
+
"loss": 0.3436,
|
| 937 |
+
"rewards/accuracies": 0.8125,
|
| 938 |
+
"rewards/chosen": -1.9234256744384766,
|
| 939 |
+
"rewards/margins": 1.9092767238616943,
|
| 940 |
+
"rewards/rejected": -3.832702159881592,
|
| 941 |
+
"step": 295
|
| 942 |
},
|
| 943 |
{
|
| 944 |
+
"epoch": 0.79,
|
| 945 |
+
"grad_norm": 52.02157401864257,
|
| 946 |
+
"learning_rate": 6.334300807088508e-08,
|
| 947 |
+
"logits/chosen": -4.835855007171631,
|
| 948 |
+
"logits/rejected": -5.021653652191162,
|
| 949 |
+
"logps/chosen": -518.7296142578125,
|
| 950 |
+
"logps/rejected": -639.1578369140625,
|
| 951 |
+
"loss": 0.3689,
|
| 952 |
+
"rewards/accuracies": 0.824999988079071,
|
| 953 |
+
"rewards/chosen": -2.530418634414673,
|
| 954 |
+
"rewards/margins": 1.8332151174545288,
|
| 955 |
+
"rewards/rejected": -4.363633155822754,
|
|
|
|
| 956 |
"step": 300
|
| 957 |
},
|
| 958 |
{
|
| 959 |
"epoch": 0.79,
|
| 960 |
+
"eval_logits/chosen": -4.618500709533691,
|
| 961 |
+
"eval_logits/rejected": -4.810915946960449,
|
| 962 |
+
"eval_logps/chosen": -895.0606079101562,
|
| 963 |
+
"eval_logps/rejected": -917.572265625,
|
| 964 |
+
"eval_loss": 1.0593960285186768,
|
| 965 |
+
"eval_rewards/accuracies": 0.5625,
|
| 966 |
+
"eval_rewards/chosen": -5.156832218170166,
|
| 967 |
+
"eval_rewards/margins": 0.2927757203578949,
|
| 968 |
+
"eval_rewards/rejected": -5.449607849121094,
|
| 969 |
+
"eval_runtime": 97.5523,
|
| 970 |
+
"eval_samples_per_second": 20.502,
|
| 971 |
+
"eval_steps_per_second": 0.328,
|
| 972 |
+
"step": 300
|
| 973 |
+
},
|
| 974 |
+
{
|
| 975 |
+
"epoch": 0.8,
|
| 976 |
+
"grad_norm": 50.197907161154795,
|
| 977 |
+
"learning_rate": 5.588272591397336e-08,
|
| 978 |
+
"logits/chosen": -4.786068916320801,
|
| 979 |
+
"logits/rejected": -4.9945068359375,
|
| 980 |
+
"logps/chosen": -485.31280517578125,
|
| 981 |
+
"logps/rejected": -626.9613647460938,
|
| 982 |
+
"loss": 0.3457,
|
| 983 |
+
"rewards/accuracies": 0.824999988079071,
|
| 984 |
+
"rewards/chosen": -2.227161169052124,
|
| 985 |
+
"rewards/margins": 2.0316004753112793,
|
| 986 |
+
"rewards/rejected": -4.258761405944824,
|
| 987 |
+
"step": 305
|
| 988 |
},
|
| 989 |
{
|
| 990 |
"epoch": 0.82,
|
| 991 |
+
"grad_norm": 48.34581690217935,
|
| 992 |
+
"learning_rate": 4.8834274139883084e-08,
|
| 993 |
+
"logits/chosen": -4.855401039123535,
|
| 994 |
+
"logits/rejected": -5.10861873626709,
|
| 995 |
+
"logps/chosen": -469.53704833984375,
|
| 996 |
+
"logps/rejected": -616.6036376953125,
|
| 997 |
+
"loss": 0.3427,
|
| 998 |
+
"rewards/accuracies": 0.8374999761581421,
|
| 999 |
+
"rewards/chosen": -2.20617413520813,
|
| 1000 |
+
"rewards/margins": 1.9504497051239014,
|
| 1001 |
+
"rewards/rejected": -4.156623840332031,
|
| 1002 |
+
"step": 310
|
| 1003 |
+
},
|
| 1004 |
+
{
|
| 1005 |
+
"epoch": 0.83,
|
| 1006 |
+
"grad_norm": 43.92355028448363,
|
| 1007 |
+
"learning_rate": 4.221260642342786e-08,
|
| 1008 |
+
"logits/chosen": -4.848982810974121,
|
| 1009 |
+
"logits/rejected": -4.972110271453857,
|
| 1010 |
+
"logps/chosen": -502.3720703125,
|
| 1011 |
+
"logps/rejected": -609.923828125,
|
| 1012 |
+
"loss": 0.3357,
|
| 1013 |
+
"rewards/accuracies": 0.8500000238418579,
|
| 1014 |
+
"rewards/chosen": -2.262599229812622,
|
| 1015 |
+
"rewards/margins": 1.6350996494293213,
|
| 1016 |
+
"rewards/rejected": -3.8976986408233643,
|
| 1017 |
+
"step": 315
|
| 1018 |
},
|
| 1019 |
{
|
| 1020 |
"epoch": 0.84,
|
| 1021 |
+
"grad_norm": 48.906535443168416,
|
| 1022 |
+
"learning_rate": 3.60317709937693e-08,
|
| 1023 |
+
"logits/chosen": -4.723662853240967,
|
| 1024 |
+
"logits/rejected": -4.999013900756836,
|
| 1025 |
+
"logps/chosen": -511.27691650390625,
|
| 1026 |
+
"logps/rejected": -656.3660888671875,
|
| 1027 |
+
"loss": 0.361,
|
| 1028 |
+
"rewards/accuracies": 0.7250000238418579,
|
| 1029 |
+
"rewards/chosen": -2.2201476097106934,
|
| 1030 |
+
"rewards/margins": 2.0478897094726562,
|
| 1031 |
+
"rewards/rejected": -4.26803731918335,
|
| 1032 |
+
"step": 320
|
| 1033 |
+
},
|
| 1034 |
+
{
|
| 1035 |
+
"epoch": 0.86,
|
| 1036 |
+
"grad_norm": 38.024316440823085,
|
| 1037 |
+
"learning_rate": 3.030488083033273e-08,
|
| 1038 |
+
"logits/chosen": -4.8308892250061035,
|
| 1039 |
+
"logits/rejected": -5.002086162567139,
|
| 1040 |
+
"logps/chosen": -500.25579833984375,
|
| 1041 |
+
"logps/rejected": -703.7189331054688,
|
| 1042 |
+
"loss": 0.3434,
|
| 1043 |
+
"rewards/accuracies": 0.8999999761581421,
|
| 1044 |
+
"rewards/chosen": -2.2706263065338135,
|
| 1045 |
+
"rewards/margins": 2.5099949836730957,
|
| 1046 |
+
"rewards/rejected": -4.780620574951172,
|
| 1047 |
+
"step": 325
|
| 1048 |
},
|
| 1049 |
{
|
| 1050 |
"epoch": 0.87,
|
| 1051 |
+
"grad_norm": 46.39394071999461,
|
| 1052 |
+
"learning_rate": 2.5044085842905683e-08,
|
| 1053 |
+
"logits/chosen": -4.880900859832764,
|
| 1054 |
+
"logits/rejected": -5.102107048034668,
|
| 1055 |
+
"logps/chosen": -522.2579345703125,
|
| 1056 |
+
"logps/rejected": -679.2675170898438,
|
| 1057 |
+
"loss": 0.375,
|
| 1058 |
+
"rewards/accuracies": 0.875,
|
| 1059 |
+
"rewards/chosen": -2.569988965988159,
|
| 1060 |
+
"rewards/margins": 2.0331876277923584,
|
| 1061 |
+
"rewards/rejected": -4.603176593780518,
|
| 1062 |
+
"step": 330
|
| 1063 |
+
},
|
| 1064 |
+
{
|
| 1065 |
+
"epoch": 0.88,
|
| 1066 |
+
"grad_norm": 44.94685644229088,
|
| 1067 |
+
"learning_rate": 2.0260547094942348e-08,
|
| 1068 |
+
"logits/chosen": -4.779486656188965,
|
| 1069 |
+
"logits/rejected": -4.989696502685547,
|
| 1070 |
+
"logps/chosen": -506.99920654296875,
|
| 1071 |
+
"logps/rejected": -654.0205688476562,
|
| 1072 |
+
"loss": 0.3649,
|
| 1073 |
+
"rewards/accuracies": 0.8500000238418579,
|
| 1074 |
+
"rewards/chosen": -2.142956495285034,
|
| 1075 |
+
"rewards/margins": 2.1117589473724365,
|
| 1076 |
+
"rewards/rejected": -4.254715919494629,
|
| 1077 |
+
"step": 335
|
| 1078 |
+
},
|
| 1079 |
+
{
|
| 1080 |
+
"epoch": 0.9,
|
| 1081 |
+
"grad_norm": 44.46821814832556,
|
| 1082 |
+
"learning_rate": 1.5964413124758493e-08,
|
| 1083 |
+
"logits/chosen": -4.752711296081543,
|
| 1084 |
+
"logits/rejected": -4.939455986022949,
|
| 1085 |
+
"logps/chosen": -438.1668395996094,
|
| 1086 |
+
"logps/rejected": -598.4129028320312,
|
| 1087 |
+
"loss": 0.3357,
|
| 1088 |
"rewards/accuracies": 0.8374999761581421,
|
| 1089 |
+
"rewards/chosen": -1.8868913650512695,
|
| 1090 |
+
"rewards/margins": 1.9845082759857178,
|
| 1091 |
+
"rewards/rejected": -3.8714001178741455,
|
| 1092 |
"step": 340
|
| 1093 |
},
|
| 1094 |
{
|
| 1095 |
+
"epoch": 0.91,
|
| 1096 |
+
"grad_norm": 49.902887693125585,
|
| 1097 |
+
"learning_rate": 1.2164798414854071e-08,
|
| 1098 |
+
"logits/chosen": -4.861344814300537,
|
| 1099 |
+
"logits/rejected": -4.92690896987915,
|
| 1100 |
+
"logps/chosen": -514.7852783203125,
|
| 1101 |
+
"logps/rejected": -692.4171142578125,
|
| 1102 |
+
"loss": 0.3343,
|
| 1103 |
+
"rewards/accuracies": 0.8125,
|
| 1104 |
+
"rewards/chosen": -2.289367437362671,
|
| 1105 |
+
"rewards/margins": 2.1290056705474854,
|
| 1106 |
+
"rewards/rejected": -4.418373107910156,
|
| 1107 |
+
"step": 345
|
| 1108 |
},
|
| 1109 |
{
|
| 1110 |
"epoch": 0.92,
|
| 1111 |
+
"grad_norm": 46.82416143070829,
|
| 1112 |
+
"learning_rate": 8.869764055041501e-09,
|
| 1113 |
+
"logits/chosen": -4.845822811126709,
|
| 1114 |
+
"logits/rejected": -5.040514945983887,
|
| 1115 |
+
"logps/chosen": -476.73358154296875,
|
| 1116 |
+
"logps/rejected": -638.4616088867188,
|
| 1117 |
+
"loss": 0.3795,
|
| 1118 |
+
"rewards/accuracies": 0.75,
|
| 1119 |
+
"rewards/chosen": -2.128450870513916,
|
| 1120 |
+
"rewards/margins": 1.8921457529067993,
|
| 1121 |
+
"rewards/rejected": -4.020596504211426,
|
| 1122 |
+
"step": 350
|
| 1123 |
+
},
|
| 1124 |
+
{
|
| 1125 |
+
"epoch": 0.94,
|
| 1126 |
+
"grad_norm": 45.59470033908034,
|
| 1127 |
+
"learning_rate": 6.086300640404079e-09,
|
| 1128 |
+
"logits/chosen": -4.7604827880859375,
|
| 1129 |
+
"logits/rejected": -4.900928020477295,
|
| 1130 |
+
"logps/chosen": -530.9966430664062,
|
| 1131 |
+
"logps/rejected": -647.1216430664062,
|
| 1132 |
+
"loss": 0.3474,
|
| 1133 |
+
"rewards/accuracies": 0.737500011920929,
|
| 1134 |
+
"rewards/chosen": -2.3973419666290283,
|
| 1135 |
+
"rewards/margins": 1.63766610622406,
|
| 1136 |
+
"rewards/rejected": -4.035007953643799,
|
| 1137 |
+
"step": 355
|
| 1138 |
},
|
| 1139 |
{
|
| 1140 |
"epoch": 0.95,
|
| 1141 |
+
"grad_norm": 38.33546651783018,
|
| 1142 |
+
"learning_rate": 3.82031344036729e-09,
|
| 1143 |
+
"logits/chosen": -4.72461462020874,
|
| 1144 |
+
"logits/rejected": -4.975947380065918,
|
| 1145 |
+
"logps/chosen": -490.10650634765625,
|
| 1146 |
+
"logps/rejected": -617.1689453125,
|
| 1147 |
+
"loss": 0.3241,
|
| 1148 |
+
"rewards/accuracies": 0.8374999761581421,
|
| 1149 |
+
"rewards/chosen": -2.2612645626068115,
|
| 1150 |
+
"rewards/margins": 1.7519699335098267,
|
| 1151 |
+
"rewards/rejected": -4.013234615325928,
|
| 1152 |
+
"step": 360
|
| 1153 |
},
|
| 1154 |
{
|
| 1155 |
+
"epoch": 0.96,
|
| 1156 |
+
"grad_norm": 48.183739221855284,
|
| 1157 |
+
"learning_rate": 2.0766098703477173e-09,
|
| 1158 |
+
"logits/chosen": -4.829585552215576,
|
| 1159 |
+
"logits/rejected": -5.007233619689941,
|
| 1160 |
+
"logps/chosen": -458.9842834472656,
|
| 1161 |
+
"logps/rejected": -589.1207275390625,
|
| 1162 |
+
"loss": 0.3698,
|
| 1163 |
+
"rewards/accuracies": 0.800000011920929,
|
| 1164 |
+
"rewards/chosen": -1.9293123483657837,
|
| 1165 |
+
"rewards/margins": 1.888193130493164,
|
| 1166 |
+
"rewards/rejected": -3.8175055980682373,
|
| 1167 |
+
"step": 365
|
| 1168 |
},
|
| 1169 |
{
|
| 1170 |
+
"epoch": 0.97,
|
| 1171 |
+
"grad_norm": 47.43892343620843,
|
| 1172 |
+
"learning_rate": 8.588892925590063e-10,
|
| 1173 |
+
"logits/chosen": -4.886306285858154,
|
| 1174 |
+
"logits/rejected": -5.221497535705566,
|
| 1175 |
+
"logps/chosen": -480.17462158203125,
|
| 1176 |
+
"logps/rejected": -637.3043823242188,
|
| 1177 |
+
"loss": 0.3276,
|
| 1178 |
"rewards/accuracies": 0.862500011920929,
|
| 1179 |
+
"rewards/chosen": -2.0167715549468994,
|
| 1180 |
+
"rewards/margins": 2.319883346557617,
|
| 1181 |
+
"rewards/rejected": -4.3366546630859375,
|
| 1182 |
+
"step": 370
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"epoch": 0.99,
|
| 1186 |
+
"grad_norm": 48.471497735173386,
|
| 1187 |
+
"learning_rate": 1.6973516761317752e-10,
|
| 1188 |
+
"logits/chosen": -4.912912845611572,
|
| 1189 |
+
"logits/rejected": -5.170907020568848,
|
| 1190 |
+
"logps/chosen": -459.01531982421875,
|
| 1191 |
+
"logps/rejected": -576.1976318359375,
|
| 1192 |
+
"loss": 0.3505,
|
| 1193 |
+
"rewards/accuracies": 0.8374999761581421,
|
| 1194 |
+
"rewards/chosen": -2.2589094638824463,
|
| 1195 |
+
"rewards/margins": 1.6588354110717773,
|
| 1196 |
+
"rewards/rejected": -3.9177448749542236,
|
| 1197 |
+
"step": 375
|
| 1198 |
},
|
| 1199 |
{
|
| 1200 |
"epoch": 1.0,
|
| 1201 |
+
"step": 379,
|
| 1202 |
"total_flos": 0.0,
|
| 1203 |
+
"train_loss": 0.4399322837512537,
|
| 1204 |
+
"train_runtime": 5833.2156,
|
| 1205 |
+
"train_samples_per_second": 8.32,
|
| 1206 |
"train_steps_per_second": 0.065
|
| 1207 |
}
|
| 1208 |
],
|
| 1209 |
+
"logging_steps": 5,
|
| 1210 |
+
"max_steps": 379,
|
| 1211 |
"num_input_tokens_seen": 0,
|
| 1212 |
"num_train_epochs": 1,
|
| 1213 |
"save_steps": 100,
|
training_args.bin
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 sha256:0a8f2cc464d4b770ce3bbadf1ed05784c3d52b8d2d804db4724e09087c1384f2
|
| 3 |
+
size 6264
|