End of training
Browse files- README.md +12 -11
- all_results.json +20 -0
- eval_results.json +15 -0
- train_results.json +8 -0
- trainer_state.json +1058 -0
- training_eval_loss.png +0 -0
- training_loss.png +0 -0
- training_rewards_accuracies.png +0 -0
README.md
CHANGED
|
@@ -3,9 +3,10 @@ library_name: peft
|
|
| 3 |
license: llama3.1
|
| 4 |
base_model: meta-llama/Llama-3.1-8B-Instruct
|
| 5 |
tags:
|
|
|
|
|
|
|
| 6 |
- trl
|
| 7 |
- dpo
|
| 8 |
-
- llama-factory
|
| 9 |
- generated_from_trainer
|
| 10 |
model-index:
|
| 11 |
- name: Llama-3.1-8B-Instruct-dpo-llama-1000
|
|
@@ -17,17 +18,17 @@ should probably proofread and complete it, then remove this comment. -->
|
|
| 17 |
|
| 18 |
# Llama-3.1-8B-Instruct-dpo-llama-1000
|
| 19 |
|
| 20 |
-
This model is a fine-tuned version of [meta-llama/Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct) on
|
| 21 |
It achieves the following results on the evaluation set:
|
| 22 |
-
- Loss: 0.
|
| 23 |
-
- Rewards/chosen: 1.
|
| 24 |
-
- Rewards/rejected: -
|
| 25 |
-
- Rewards/accuracies: 0.
|
| 26 |
-
- Rewards/margins:
|
| 27 |
-
- Logps/chosen: -
|
| 28 |
-
- Logps/rejected: -
|
| 29 |
-
- Logits/chosen: -0.
|
| 30 |
-
- Logits/rejected: -0.
|
| 31 |
|
| 32 |
## Model description
|
| 33 |
|
|
|
|
| 3 |
license: llama3.1
|
| 4 |
base_model: meta-llama/Llama-3.1-8B-Instruct
|
| 5 |
tags:
|
| 6 |
+
- llama-factory
|
| 7 |
+
- lora
|
| 8 |
- trl
|
| 9 |
- dpo
|
|
|
|
| 10 |
- generated_from_trainer
|
| 11 |
model-index:
|
| 12 |
- name: Llama-3.1-8B-Instruct-dpo-llama-1000
|
|
|
|
| 18 |
|
| 19 |
# Llama-3.1-8B-Instruct-dpo-llama-1000
|
| 20 |
|
| 21 |
+
This model is a fine-tuned version of [meta-llama/Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct) on the answer_llama dataset.
|
| 22 |
It achieves the following results on the evaluation set:
|
| 23 |
+
- Loss: 0.3077
|
| 24 |
+
- Rewards/chosen: 1.4814
|
| 25 |
+
- Rewards/rejected: -0.7600
|
| 26 |
+
- Rewards/accuracies: 0.8500
|
| 27 |
+
- Rewards/margins: 2.2414
|
| 28 |
+
- Logps/chosen: -7.6796
|
| 29 |
+
- Logps/rejected: -31.9936
|
| 30 |
+
- Logits/chosen: -0.2154
|
| 31 |
+
- Logits/rejected: -0.3106
|
| 32 |
|
| 33 |
## Model description
|
| 34 |
|
all_results.json
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 9.955555555555556,
|
| 3 |
+
"eval_logits/chosen": -0.21535295248031616,
|
| 4 |
+
"eval_logits/rejected": -0.31062132120132446,
|
| 5 |
+
"eval_logps/chosen": -7.679609775543213,
|
| 6 |
+
"eval_logps/rejected": -31.993576049804688,
|
| 7 |
+
"eval_loss": 0.307707816362381,
|
| 8 |
+
"eval_rewards/accuracies": 0.8499999642372131,
|
| 9 |
+
"eval_rewards/chosen": 1.48140287399292,
|
| 10 |
+
"eval_rewards/margins": 2.2414333820343018,
|
| 11 |
+
"eval_rewards/rejected": -0.7600305080413818,
|
| 12 |
+
"eval_runtime": 12.0737,
|
| 13 |
+
"eval_samples_per_second": 8.282,
|
| 14 |
+
"eval_steps_per_second": 4.141,
|
| 15 |
+
"total_flos": 6.741083695664333e+16,
|
| 16 |
+
"train_loss": 0.31093634622437616,
|
| 17 |
+
"train_runtime": 2833.3873,
|
| 18 |
+
"train_samples_per_second": 3.176,
|
| 19 |
+
"train_steps_per_second": 0.198
|
| 20 |
+
}
|
eval_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 9.955555555555556,
|
| 3 |
+
"eval_logits/chosen": -0.21535295248031616,
|
| 4 |
+
"eval_logits/rejected": -0.31062132120132446,
|
| 5 |
+
"eval_logps/chosen": -7.679609775543213,
|
| 6 |
+
"eval_logps/rejected": -31.993576049804688,
|
| 7 |
+
"eval_loss": 0.307707816362381,
|
| 8 |
+
"eval_rewards/accuracies": 0.8499999642372131,
|
| 9 |
+
"eval_rewards/chosen": 1.48140287399292,
|
| 10 |
+
"eval_rewards/margins": 2.2414333820343018,
|
| 11 |
+
"eval_rewards/rejected": -0.7600305080413818,
|
| 12 |
+
"eval_runtime": 12.0737,
|
| 13 |
+
"eval_samples_per_second": 8.282,
|
| 14 |
+
"eval_steps_per_second": 4.141
|
| 15 |
+
}
|
train_results.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 9.955555555555556,
|
| 3 |
+
"total_flos": 6.741083695664333e+16,
|
| 4 |
+
"train_loss": 0.31093634622437616,
|
| 5 |
+
"train_runtime": 2833.3873,
|
| 6 |
+
"train_samples_per_second": 3.176,
|
| 7 |
+
"train_steps_per_second": 0.198
|
| 8 |
+
}
|
trainer_state.json
ADDED
|
@@ -0,0 +1,1058 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": 0.307707816362381,
|
| 3 |
+
"best_model_checkpoint": "saves/sycophancy/Llama-3.1-8B-Instruct/dpo-llama-1000/train/checkpoint-200",
|
| 4 |
+
"epoch": 9.955555555555556,
|
| 5 |
+
"eval_steps": 50,
|
| 6 |
+
"global_step": 560,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.17777777777777778,
|
| 13 |
+
"grad_norm": 1.8430479764938354,
|
| 14 |
+
"learning_rate": 8.928571428571429e-07,
|
| 15 |
+
"logits/chosen": -0.42922043800354004,
|
| 16 |
+
"logits/rejected": -0.48287302255630493,
|
| 17 |
+
"logps/chosen": -22.499736785888672,
|
| 18 |
+
"logps/rejected": -25.05568504333496,
|
| 19 |
+
"loss": 0.693,
|
| 20 |
+
"rewards/accuracies": 0.4312500059604645,
|
| 21 |
+
"rewards/chosen": 3.8243924791458994e-05,
|
| 22 |
+
"rewards/margins": 0.00029587256722152233,
|
| 23 |
+
"rewards/rejected": -0.00025762844597920775,
|
| 24 |
+
"step": 10
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.35555555555555557,
|
| 28 |
+
"grad_norm": 1.1751757860183716,
|
| 29 |
+
"learning_rate": 1.7857142857142859e-06,
|
| 30 |
+
"logits/chosen": -0.4168773591518402,
|
| 31 |
+
"logits/rejected": -0.4500039517879486,
|
| 32 |
+
"logps/chosen": -23.05335807800293,
|
| 33 |
+
"logps/rejected": -24.607397079467773,
|
| 34 |
+
"loss": 0.6923,
|
| 35 |
+
"rewards/accuracies": 0.543749988079071,
|
| 36 |
+
"rewards/chosen": 0.0031167108099907637,
|
| 37 |
+
"rewards/margins": 0.001834285445511341,
|
| 38 |
+
"rewards/rejected": 0.0012824248988181353,
|
| 39 |
+
"step": 20
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.5333333333333333,
|
| 43 |
+
"grad_norm": 2.2308878898620605,
|
| 44 |
+
"learning_rate": 2.6785714285714285e-06,
|
| 45 |
+
"logits/chosen": -0.4415621757507324,
|
| 46 |
+
"logits/rejected": -0.47939205169677734,
|
| 47 |
+
"logps/chosen": -23.05750846862793,
|
| 48 |
+
"logps/rejected": -24.027509689331055,
|
| 49 |
+
"loss": 0.6925,
|
| 50 |
+
"rewards/accuracies": 0.4937500059604645,
|
| 51 |
+
"rewards/chosen": 0.006902683060616255,
|
| 52 |
+
"rewards/margins": 0.0014179922873154283,
|
| 53 |
+
"rewards/rejected": 0.005484690889716148,
|
| 54 |
+
"step": 30
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"epoch": 0.7111111111111111,
|
| 58 |
+
"grad_norm": 1.4753832817077637,
|
| 59 |
+
"learning_rate": 3.5714285714285718e-06,
|
| 60 |
+
"logits/chosen": -0.4203382134437561,
|
| 61 |
+
"logits/rejected": -0.4865742623806,
|
| 62 |
+
"logps/chosen": -21.71892738342285,
|
| 63 |
+
"logps/rejected": -24.279401779174805,
|
| 64 |
+
"loss": 0.6879,
|
| 65 |
+
"rewards/accuracies": 0.6312500238418579,
|
| 66 |
+
"rewards/chosen": 0.019825046882033348,
|
| 67 |
+
"rewards/margins": 0.010711194016039371,
|
| 68 |
+
"rewards/rejected": 0.009113854728639126,
|
| 69 |
+
"step": 40
|
| 70 |
+
},
|
| 71 |
+
{
|
| 72 |
+
"epoch": 0.8888888888888888,
|
| 73 |
+
"grad_norm": 1.723747968673706,
|
| 74 |
+
"learning_rate": 4.464285714285715e-06,
|
| 75 |
+
"logits/chosen": -0.45647165179252625,
|
| 76 |
+
"logits/rejected": -0.4898137152194977,
|
| 77 |
+
"logps/chosen": -22.14567756652832,
|
| 78 |
+
"logps/rejected": -24.119258880615234,
|
| 79 |
+
"loss": 0.6815,
|
| 80 |
+
"rewards/accuracies": 0.6312500238418579,
|
| 81 |
+
"rewards/chosen": 0.04727761819958687,
|
| 82 |
+
"rewards/margins": 0.024313444271683693,
|
| 83 |
+
"rewards/rejected": 0.022964173927903175,
|
| 84 |
+
"step": 50
|
| 85 |
+
},
|
| 86 |
+
{
|
| 87 |
+
"epoch": 0.8888888888888888,
|
| 88 |
+
"eval_logits/chosen": -0.4114196300506592,
|
| 89 |
+
"eval_logits/rejected": -0.4791559875011444,
|
| 90 |
+
"eval_logps/chosen": -21.660146713256836,
|
| 91 |
+
"eval_logps/rejected": -24.039798736572266,
|
| 92 |
+
"eval_loss": 0.6706590056419373,
|
| 93 |
+
"eval_rewards/accuracies": 0.6899999976158142,
|
| 94 |
+
"eval_rewards/chosen": 0.08334928005933762,
|
| 95 |
+
"eval_rewards/margins": 0.048002347350120544,
|
| 96 |
+
"eval_rewards/rejected": 0.03534693643450737,
|
| 97 |
+
"eval_runtime": 12.9721,
|
| 98 |
+
"eval_samples_per_second": 7.709,
|
| 99 |
+
"eval_steps_per_second": 3.854,
|
| 100 |
+
"step": 50
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 1.0666666666666667,
|
| 104 |
+
"grad_norm": 1.8534899950027466,
|
| 105 |
+
"learning_rate": 4.999222955002041e-06,
|
| 106 |
+
"logits/chosen": -0.4518943428993225,
|
| 107 |
+
"logits/rejected": -0.4910917282104492,
|
| 108 |
+
"logps/chosen": -22.00179672241211,
|
| 109 |
+
"logps/rejected": -23.545581817626953,
|
| 110 |
+
"loss": 0.6705,
|
| 111 |
+
"rewards/accuracies": 0.6312500238418579,
|
| 112 |
+
"rewards/chosen": 0.106198251247406,
|
| 113 |
+
"rewards/margins": 0.05078822001814842,
|
| 114 |
+
"rewards/rejected": 0.05541003867983818,
|
| 115 |
+
"step": 60
|
| 116 |
+
},
|
| 117 |
+
{
|
| 118 |
+
"epoch": 1.2444444444444445,
|
| 119 |
+
"grad_norm": 2.505237340927124,
|
| 120 |
+
"learning_rate": 4.990486745229364e-06,
|
| 121 |
+
"logits/chosen": -0.42713257670402527,
|
| 122 |
+
"logits/rejected": -0.48525214195251465,
|
| 123 |
+
"logps/chosen": -20.35358428955078,
|
| 124 |
+
"logps/rejected": -23.495656967163086,
|
| 125 |
+
"loss": 0.6328,
|
| 126 |
+
"rewards/accuracies": 0.6875,
|
| 127 |
+
"rewards/chosen": 0.25575509667396545,
|
| 128 |
+
"rewards/margins": 0.14482033252716064,
|
| 129 |
+
"rewards/rejected": 0.11093475669622421,
|
| 130 |
+
"step": 70
|
| 131 |
+
},
|
| 132 |
+
{
|
| 133 |
+
"epoch": 1.4222222222222223,
|
| 134 |
+
"grad_norm": 2.34883713722229,
|
| 135 |
+
"learning_rate": 4.9720770655628216e-06,
|
| 136 |
+
"logits/chosen": -0.40827736258506775,
|
| 137 |
+
"logits/rejected": -0.45998048782348633,
|
| 138 |
+
"logps/chosen": -19.513925552368164,
|
| 139 |
+
"logps/rejected": -23.603103637695312,
|
| 140 |
+
"loss": 0.5899,
|
| 141 |
+
"rewards/accuracies": 0.7250000238418579,
|
| 142 |
+
"rewards/chosen": 0.4022657573223114,
|
| 143 |
+
"rewards/margins": 0.26808053255081177,
|
| 144 |
+
"rewards/rejected": 0.13418518006801605,
|
| 145 |
+
"step": 80
|
| 146 |
+
},
|
| 147 |
+
{
|
| 148 |
+
"epoch": 1.6,
|
| 149 |
+
"grad_norm": 3.0297181606292725,
|
| 150 |
+
"learning_rate": 4.944065422298262e-06,
|
| 151 |
+
"logits/chosen": -0.3988552689552307,
|
| 152 |
+
"logits/rejected": -0.465530127286911,
|
| 153 |
+
"logps/chosen": -14.463418960571289,
|
| 154 |
+
"logps/rejected": -21.68548583984375,
|
| 155 |
+
"loss": 0.5208,
|
| 156 |
+
"rewards/accuracies": 0.7750000357627869,
|
| 157 |
+
"rewards/chosen": 0.7595838904380798,
|
| 158 |
+
"rewards/margins": 0.5038682818412781,
|
| 159 |
+
"rewards/rejected": 0.25571563839912415,
|
| 160 |
+
"step": 90
|
| 161 |
+
},
|
| 162 |
+
{
|
| 163 |
+
"epoch": 1.7777777777777777,
|
| 164 |
+
"grad_norm": 2.1064534187316895,
|
| 165 |
+
"learning_rate": 4.90656061737503e-06,
|
| 166 |
+
"logits/chosen": -0.3954499363899231,
|
| 167 |
+
"logits/rejected": -0.448087602853775,
|
| 168 |
+
"logps/chosen": -13.89409065246582,
|
| 169 |
+
"logps/rejected": -21.66886329650879,
|
| 170 |
+
"loss": 0.5082,
|
| 171 |
+
"rewards/accuracies": 0.6937500238418579,
|
| 172 |
+
"rewards/chosen": 0.9080438613891602,
|
| 173 |
+
"rewards/margins": 0.6023429036140442,
|
| 174 |
+
"rewards/rejected": 0.30570098757743835,
|
| 175 |
+
"step": 100
|
| 176 |
+
},
|
| 177 |
+
{
|
| 178 |
+
"epoch": 1.7777777777777777,
|
| 179 |
+
"eval_logits/chosen": -0.3558562695980072,
|
| 180 |
+
"eval_logits/rejected": -0.4376991093158722,
|
| 181 |
+
"eval_logps/chosen": -12.185461044311523,
|
| 182 |
+
"eval_logps/rejected": -22.450639724731445,
|
| 183 |
+
"eval_loss": 0.4427681863307953,
|
| 184 |
+
"eval_rewards/accuracies": 0.7899999618530273,
|
| 185 |
+
"eval_rewards/chosen": 1.0308177471160889,
|
| 186 |
+
"eval_rewards/margins": 0.8365544080734253,
|
| 187 |
+
"eval_rewards/rejected": 0.19426332414150238,
|
| 188 |
+
"eval_runtime": 12.9808,
|
| 189 |
+
"eval_samples_per_second": 7.704,
|
| 190 |
+
"eval_steps_per_second": 3.852,
|
| 191 |
+
"step": 100
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 1.9555555555555557,
|
| 195 |
+
"grad_norm": 2.931056261062622,
|
| 196 |
+
"learning_rate": 4.859708325770919e-06,
|
| 197 |
+
"logits/chosen": -0.3505342900753021,
|
| 198 |
+
"logits/rejected": -0.41251203417778015,
|
| 199 |
+
"logps/chosen": -12.102005004882812,
|
| 200 |
+
"logps/rejected": -22.603740692138672,
|
| 201 |
+
"loss": 0.4419,
|
| 202 |
+
"rewards/accuracies": 0.8375000357627869,
|
| 203 |
+
"rewards/chosen": 0.9892138838768005,
|
| 204 |
+
"rewards/margins": 0.8685030341148376,
|
| 205 |
+
"rewards/rejected": 0.12071088701486588,
|
| 206 |
+
"step": 110
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"epoch": 2.1333333333333333,
|
| 210 |
+
"grad_norm": 3.659130573272705,
|
| 211 |
+
"learning_rate": 4.80369052967602e-06,
|
| 212 |
+
"logits/chosen": -0.3336087167263031,
|
| 213 |
+
"logits/rejected": -0.4222935140132904,
|
| 214 |
+
"logps/chosen": -11.186881065368652,
|
| 215 |
+
"logps/rejected": -24.831924438476562,
|
| 216 |
+
"loss": 0.365,
|
| 217 |
+
"rewards/accuracies": 0.84375,
|
| 218 |
+
"rewards/chosen": 1.164413332939148,
|
| 219 |
+
"rewards/margins": 1.20394766330719,
|
| 220 |
+
"rewards/rejected": -0.03953445702791214,
|
| 221 |
+
"step": 120
|
| 222 |
+
},
|
| 223 |
+
{
|
| 224 |
+
"epoch": 2.311111111111111,
|
| 225 |
+
"grad_norm": 3.590104103088379,
|
| 226 |
+
"learning_rate": 4.7387248116432524e-06,
|
| 227 |
+
"logits/chosen": -0.319705069065094,
|
| 228 |
+
"logits/rejected": -0.4010167717933655,
|
| 229 |
+
"logps/chosen": -10.573092460632324,
|
| 230 |
+
"logps/rejected": -25.705326080322266,
|
| 231 |
+
"loss": 0.3523,
|
| 232 |
+
"rewards/accuracies": 0.824999988079071,
|
| 233 |
+
"rewards/chosen": 1.1774530410766602,
|
| 234 |
+
"rewards/margins": 1.3132127523422241,
|
| 235 |
+
"rewards/rejected": -0.13575978577136993,
|
| 236 |
+
"step": 130
|
| 237 |
+
},
|
| 238 |
+
{
|
| 239 |
+
"epoch": 2.488888888888889,
|
| 240 |
+
"grad_norm": 3.342087984085083,
|
| 241 |
+
"learning_rate": 4.665063509461098e-06,
|
| 242 |
+
"logits/chosen": -0.33713865280151367,
|
| 243 |
+
"logits/rejected": -0.40943965315818787,
|
| 244 |
+
"logps/chosen": -11.759323120117188,
|
| 245 |
+
"logps/rejected": -27.167184829711914,
|
| 246 |
+
"loss": 0.3881,
|
| 247 |
+
"rewards/accuracies": 0.824999988079071,
|
| 248 |
+
"rewards/chosen": 1.0846505165100098,
|
| 249 |
+
"rewards/margins": 1.4101663827896118,
|
| 250 |
+
"rewards/rejected": -0.3255158066749573,
|
| 251 |
+
"step": 140
|
| 252 |
+
},
|
| 253 |
+
{
|
| 254 |
+
"epoch": 2.6666666666666665,
|
| 255 |
+
"grad_norm": 1.9896889925003052,
|
| 256 |
+
"learning_rate": 4.5829927360311224e-06,
|
| 257 |
+
"logits/chosen": -0.2843226492404938,
|
| 258 |
+
"logits/rejected": -0.3789869546890259,
|
| 259 |
+
"logps/chosen": -10.016444206237793,
|
| 260 |
+
"logps/rejected": -29.09038734436035,
|
| 261 |
+
"loss": 0.2979,
|
| 262 |
+
"rewards/accuracies": 0.862500011920929,
|
| 263 |
+
"rewards/chosen": 1.313085675239563,
|
| 264 |
+
"rewards/margins": 1.761974573135376,
|
| 265 |
+
"rewards/rejected": -0.4488888680934906,
|
| 266 |
+
"step": 150
|
| 267 |
+
},
|
| 268 |
+
{
|
| 269 |
+
"epoch": 2.6666666666666665,
|
| 270 |
+
"eval_logits/chosen": -0.2695091664791107,
|
| 271 |
+
"eval_logits/rejected": -0.36552613973617554,
|
| 272 |
+
"eval_logps/chosen": -9.01309871673584,
|
| 273 |
+
"eval_logps/rejected": -28.563730239868164,
|
| 274 |
+
"eval_loss": 0.32149210572242737,
|
| 275 |
+
"eval_rewards/accuracies": 0.85999995470047,
|
| 276 |
+
"eval_rewards/chosen": 1.348054051399231,
|
| 277 |
+
"eval_rewards/margins": 1.7650997638702393,
|
| 278 |
+
"eval_rewards/rejected": -0.4170458912849426,
|
| 279 |
+
"eval_runtime": 13.0246,
|
| 280 |
+
"eval_samples_per_second": 7.678,
|
| 281 |
+
"eval_steps_per_second": 3.839,
|
| 282 |
+
"step": 150
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 2.8444444444444446,
|
| 286 |
+
"grad_norm": 1.8847765922546387,
|
| 287 |
+
"learning_rate": 4.492831268057307e-06,
|
| 288 |
+
"logits/chosen": -0.2392854541540146,
|
| 289 |
+
"logits/rejected": -0.32519179582595825,
|
| 290 |
+
"logps/chosen": -10.91231918334961,
|
| 291 |
+
"logps/rejected": -29.385107040405273,
|
| 292 |
+
"loss": 0.3806,
|
| 293 |
+
"rewards/accuracies": 0.8375000357627869,
|
| 294 |
+
"rewards/chosen": 1.174955129623413,
|
| 295 |
+
"rewards/margins": 1.6338344812393188,
|
| 296 |
+
"rewards/rejected": -0.458879292011261,
|
| 297 |
+
"step": 160
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"epoch": 3.022222222222222,
|
| 301 |
+
"grad_norm": 4.04591703414917,
|
| 302 |
+
"learning_rate": 4.394929307863633e-06,
|
| 303 |
+
"logits/chosen": -0.25453540682792664,
|
| 304 |
+
"logits/rejected": -0.3346460461616516,
|
| 305 |
+
"logps/chosen": -9.183024406433105,
|
| 306 |
+
"logps/rejected": -30.89423942565918,
|
| 307 |
+
"loss": 0.2928,
|
| 308 |
+
"rewards/accuracies": 0.875,
|
| 309 |
+
"rewards/chosen": 1.3575671911239624,
|
| 310 |
+
"rewards/margins": 1.941465973854065,
|
| 311 |
+
"rewards/rejected": -0.5838987231254578,
|
| 312 |
+
"step": 170
|
| 313 |
+
},
|
| 314 |
+
{
|
| 315 |
+
"epoch": 3.2,
|
| 316 |
+
"grad_norm": 4.6089653968811035,
|
| 317 |
+
"learning_rate": 4.289667123149296e-06,
|
| 318 |
+
"logits/chosen": -0.2588854432106018,
|
| 319 |
+
"logits/rejected": -0.344396710395813,
|
| 320 |
+
"logps/chosen": -8.893033027648926,
|
| 321 |
+
"logps/rejected": -30.978750228881836,
|
| 322 |
+
"loss": 0.2991,
|
| 323 |
+
"rewards/accuracies": 0.8812500238418579,
|
| 324 |
+
"rewards/chosen": 1.382232427597046,
|
| 325 |
+
"rewards/margins": 1.9790451526641846,
|
| 326 |
+
"rewards/rejected": -0.596812903881073,
|
| 327 |
+
"step": 180
|
| 328 |
+
},
|
| 329 |
+
{
|
| 330 |
+
"epoch": 3.3777777777777778,
|
| 331 |
+
"grad_norm": 2.9751718044281006,
|
| 332 |
+
"learning_rate": 4.177453569964925e-06,
|
| 333 |
+
"logits/chosen": -0.21491435170173645,
|
| 334 |
+
"logits/rejected": -0.30069640278816223,
|
| 335 |
+
"logps/chosen": -8.679551124572754,
|
| 336 |
+
"logps/rejected": -32.274330139160156,
|
| 337 |
+
"loss": 0.2848,
|
| 338 |
+
"rewards/accuracies": 0.8687500357627869,
|
| 339 |
+
"rewards/chosen": 1.3893749713897705,
|
| 340 |
+
"rewards/margins": 2.1629250049591064,
|
| 341 |
+
"rewards/rejected": -0.7735500931739807,
|
| 342 |
+
"step": 190
|
| 343 |
+
},
|
| 344 |
+
{
|
| 345 |
+
"epoch": 3.5555555555555554,
|
| 346 |
+
"grad_norm": 4.007388591766357,
|
| 347 |
+
"learning_rate": 4.058724504646834e-06,
|
| 348 |
+
"logits/chosen": -0.2135535329580307,
|
| 349 |
+
"logits/rejected": -0.280956506729126,
|
| 350 |
+
"logps/chosen": -8.478487968444824,
|
| 351 |
+
"logps/rejected": -31.442169189453125,
|
| 352 |
+
"loss": 0.2862,
|
| 353 |
+
"rewards/accuracies": 0.8500000238418579,
|
| 354 |
+
"rewards/chosen": 1.4557212591171265,
|
| 355 |
+
"rewards/margins": 2.184141159057617,
|
| 356 |
+
"rewards/rejected": -0.7284198999404907,
|
| 357 |
+
"step": 200
|
| 358 |
+
},
|
| 359 |
+
{
|
| 360 |
+
"epoch": 3.5555555555555554,
|
| 361 |
+
"eval_logits/chosen": -0.21535295248031616,
|
| 362 |
+
"eval_logits/rejected": -0.31062132120132446,
|
| 363 |
+
"eval_logps/chosen": -7.679609775543213,
|
| 364 |
+
"eval_logps/rejected": -31.993576049804688,
|
| 365 |
+
"eval_loss": 0.307707816362381,
|
| 366 |
+
"eval_rewards/accuracies": 0.8499999642372131,
|
| 367 |
+
"eval_rewards/chosen": 1.48140287399292,
|
| 368 |
+
"eval_rewards/margins": 2.2414333820343018,
|
| 369 |
+
"eval_rewards/rejected": -0.7600305080413818,
|
| 370 |
+
"eval_runtime": 13.0936,
|
| 371 |
+
"eval_samples_per_second": 7.637,
|
| 372 |
+
"eval_steps_per_second": 3.819,
|
| 373 |
+
"step": 200
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 3.7333333333333334,
|
| 377 |
+
"grad_norm": 20.840646743774414,
|
| 378 |
+
"learning_rate": 3.933941090877615e-06,
|
| 379 |
+
"logits/chosen": -0.21822166442871094,
|
| 380 |
+
"logits/rejected": -0.3239297866821289,
|
| 381 |
+
"logps/chosen": -10.312289237976074,
|
| 382 |
+
"logps/rejected": -33.20119857788086,
|
| 383 |
+
"loss": 0.3321,
|
| 384 |
+
"rewards/accuracies": 0.887499988079071,
|
| 385 |
+
"rewards/chosen": 1.287852168083191,
|
| 386 |
+
"rewards/margins": 2.1572673320770264,
|
| 387 |
+
"rewards/rejected": -0.8694152235984802,
|
| 388 |
+
"step": 210
|
| 389 |
+
},
|
| 390 |
+
{
|
| 391 |
+
"epoch": 3.911111111111111,
|
| 392 |
+
"grad_norm": 15.999884605407715,
|
| 393 |
+
"learning_rate": 3.8035880084487454e-06,
|
| 394 |
+
"logits/chosen": -0.21503937244415283,
|
| 395 |
+
"logits/rejected": -0.29350587725639343,
|
| 396 |
+
"logps/chosen": -8.648448944091797,
|
| 397 |
+
"logps/rejected": -34.50216293334961,
|
| 398 |
+
"loss": 0.2686,
|
| 399 |
+
"rewards/accuracies": 0.893750011920929,
|
| 400 |
+
"rewards/chosen": 1.3622721433639526,
|
| 401 |
+
"rewards/margins": 2.405822992324829,
|
| 402 |
+
"rewards/rejected": -1.0435508489608765,
|
| 403 |
+
"step": 220
|
| 404 |
+
},
|
| 405 |
+
{
|
| 406 |
+
"epoch": 4.088888888888889,
|
| 407 |
+
"grad_norm": 7.8126959800720215,
|
| 408 |
+
"learning_rate": 3.6681715706826555e-06,
|
| 409 |
+
"logits/chosen": -0.22714447975158691,
|
| 410 |
+
"logits/rejected": -0.3037336468696594,
|
| 411 |
+
"logps/chosen": -9.144305229187012,
|
| 412 |
+
"logps/rejected": -35.50819778442383,
|
| 413 |
+
"loss": 0.2284,
|
| 414 |
+
"rewards/accuracies": 0.9125000238418579,
|
| 415 |
+
"rewards/chosen": 1.3389980792999268,
|
| 416 |
+
"rewards/margins": 2.4519691467285156,
|
| 417 |
+
"rewards/rejected": -1.1129711866378784,
|
| 418 |
+
"step": 230
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 4.266666666666667,
|
| 422 |
+
"grad_norm": 9.81760311126709,
|
| 423 |
+
"learning_rate": 3.5282177578265295e-06,
|
| 424 |
+
"logits/chosen": -0.1930251568555832,
|
| 425 |
+
"logits/rejected": -0.28283536434173584,
|
| 426 |
+
"logps/chosen": -9.54443073272705,
|
| 427 |
+
"logps/rejected": -36.10524368286133,
|
| 428 |
+
"loss": 0.2864,
|
| 429 |
+
"rewards/accuracies": 0.9000000357627869,
|
| 430 |
+
"rewards/chosen": 1.4175978899002075,
|
| 431 |
+
"rewards/margins": 2.55643630027771,
|
| 432 |
+
"rewards/rejected": -1.1388384103775024,
|
| 433 |
+
"step": 240
|
| 434 |
+
},
|
| 435 |
+
{
|
| 436 |
+
"epoch": 4.444444444444445,
|
| 437 |
+
"grad_norm": 5.866613864898682,
|
| 438 |
+
"learning_rate": 3.384270174056454e-06,
|
| 439 |
+
"logits/chosen": -0.21315816044807434,
|
| 440 |
+
"logits/rejected": -0.30596357583999634,
|
| 441 |
+
"logps/chosen": -8.75170612335205,
|
| 442 |
+
"logps/rejected": -37.464111328125,
|
| 443 |
+
"loss": 0.2747,
|
| 444 |
+
"rewards/accuracies": 0.8812500238418579,
|
| 445 |
+
"rewards/chosen": 1.375083088874817,
|
| 446 |
+
"rewards/margins": 2.6868672370910645,
|
| 447 |
+
"rewards/rejected": -1.3117841482162476,
|
| 448 |
+
"step": 250
|
| 449 |
+
},
|
| 450 |
+
{
|
| 451 |
+
"epoch": 4.444444444444445,
|
| 452 |
+
"eval_logits/chosen": -0.18717029690742493,
|
| 453 |
+
"eval_logits/rejected": -0.28792956471443176,
|
| 454 |
+
"eval_logps/chosen": -8.346588134765625,
|
| 455 |
+
"eval_logps/rejected": -36.838497161865234,
|
| 456 |
+
"eval_loss": 0.3183891177177429,
|
| 457 |
+
"eval_rewards/accuracies": 0.85999995470047,
|
| 458 |
+
"eval_rewards/chosen": 1.4147050380706787,
|
| 459 |
+
"eval_rewards/margins": 2.6592278480529785,
|
| 460 |
+
"eval_rewards/rejected": -1.2445228099822998,
|
| 461 |
+
"eval_runtime": 13.1818,
|
| 462 |
+
"eval_samples_per_second": 7.586,
|
| 463 |
+
"eval_steps_per_second": 3.793,
|
| 464 |
+
"step": 250
|
| 465 |
+
},
|
| 466 |
+
{
|
| 467 |
+
"epoch": 4.622222222222222,
|
| 468 |
+
"grad_norm": 1.7029662132263184,
|
| 469 |
+
"learning_rate": 3.236887936027261e-06,
|
| 470 |
+
"logits/chosen": -0.19846662878990173,
|
| 471 |
+
"logits/rejected": -0.27518361806869507,
|
| 472 |
+
"logps/chosen": -8.800247192382812,
|
| 473 |
+
"logps/rejected": -38.98476791381836,
|
| 474 |
+
"loss": 0.2502,
|
| 475 |
+
"rewards/accuracies": 0.862500011920929,
|
| 476 |
+
"rewards/chosen": 1.383112907409668,
|
| 477 |
+
"rewards/margins": 2.8278017044067383,
|
| 478 |
+
"rewards/rejected": -1.4446886777877808,
|
| 479 |
+
"step": 260
|
| 480 |
+
},
|
| 481 |
+
{
|
| 482 |
+
"epoch": 4.8,
|
| 483 |
+
"grad_norm": 0.8134036064147949,
|
| 484 |
+
"learning_rate": 3.0866435011692884e-06,
|
| 485 |
+
"logits/chosen": -0.1693725436925888,
|
| 486 |
+
"logits/rejected": -0.2626606523990631,
|
| 487 |
+
"logps/chosen": -7.444068908691406,
|
| 488 |
+
"logps/rejected": -39.20288848876953,
|
| 489 |
+
"loss": 0.2289,
|
| 490 |
+
"rewards/accuracies": 0.9000000357627869,
|
| 491 |
+
"rewards/chosen": 1.527998447418213,
|
| 492 |
+
"rewards/margins": 2.998661994934082,
|
| 493 |
+
"rewards/rejected": -1.47066330909729,
|
| 494 |
+
"step": 270
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 4.977777777777778,
|
| 498 |
+
"grad_norm": 6.755254745483398,
|
| 499 |
+
"learning_rate": 2.9341204441673267e-06,
|
| 500 |
+
"logits/chosen": -0.1920958012342453,
|
| 501 |
+
"logits/rejected": -0.28297463059425354,
|
| 502 |
+
"logps/chosen": -8.827613830566406,
|
| 503 |
+
"logps/rejected": -37.988773345947266,
|
| 504 |
+
"loss": 0.2723,
|
| 505 |
+
"rewards/accuracies": 0.8812500238418579,
|
| 506 |
+
"rewards/chosen": 1.3093467950820923,
|
| 507 |
+
"rewards/margins": 2.6953580379486084,
|
| 508 |
+
"rewards/rejected": -1.3860112428665161,
|
| 509 |
+
"step": 280
|
| 510 |
+
},
|
| 511 |
+
{
|
| 512 |
+
"epoch": 5.155555555555556,
|
| 513 |
+
"grad_norm": 8.348061561584473,
|
| 514 |
+
"learning_rate": 2.7799111902582697e-06,
|
| 515 |
+
"logits/chosen": -0.19827289879322052,
|
| 516 |
+
"logits/rejected": -0.28890833258628845,
|
| 517 |
+
"logps/chosen": -8.641542434692383,
|
| 518 |
+
"logps/rejected": -37.83559036254883,
|
| 519 |
+
"loss": 0.2319,
|
| 520 |
+
"rewards/accuracies": 0.90625,
|
| 521 |
+
"rewards/chosen": 1.4176464080810547,
|
| 522 |
+
"rewards/margins": 2.750339984893799,
|
| 523 |
+
"rewards/rejected": -1.3326934576034546,
|
| 524 |
+
"step": 290
|
| 525 |
+
},
|
| 526 |
+
{
|
| 527 |
+
"epoch": 5.333333333333333,
|
| 528 |
+
"grad_norm": 4.27839469909668,
|
| 529 |
+
"learning_rate": 2.624614714151743e-06,
|
| 530 |
+
"logits/chosen": -0.16082732379436493,
|
| 531 |
+
"logits/rejected": -0.2566200792789459,
|
| 532 |
+
"logps/chosen": -7.895478248596191,
|
| 533 |
+
"logps/rejected": -37.70954132080078,
|
| 534 |
+
"loss": 0.2688,
|
| 535 |
+
"rewards/accuracies": 0.8812500238418579,
|
| 536 |
+
"rewards/chosen": 1.4443522691726685,
|
| 537 |
+
"rewards/margins": 2.8291947841644287,
|
| 538 |
+
"rewards/rejected": -1.3848422765731812,
|
| 539 |
+
"step": 300
|
| 540 |
+
},
|
| 541 |
+
{
|
| 542 |
+
"epoch": 5.333333333333333,
|
| 543 |
+
"eval_logits/chosen": -0.17137397825717926,
|
| 544 |
+
"eval_logits/rejected": -0.27050745487213135,
|
| 545 |
+
"eval_logps/chosen": -8.0242280960083,
|
| 546 |
+
"eval_logps/rejected": -37.18735122680664,
|
| 547 |
+
"eval_loss": 0.3195304572582245,
|
| 548 |
+
"eval_rewards/accuracies": 0.8499999642372131,
|
| 549 |
+
"eval_rewards/chosen": 1.4469408988952637,
|
| 550 |
+
"eval_rewards/margins": 2.726349115371704,
|
| 551 |
+
"eval_rewards/rejected": -1.2794082164764404,
|
| 552 |
+
"eval_runtime": 13.1946,
|
| 553 |
+
"eval_samples_per_second": 7.579,
|
| 554 |
+
"eval_steps_per_second": 3.789,
|
| 555 |
+
"step": 300
|
| 556 |
+
},
|
| 557 |
+
{
|
| 558 |
+
"epoch": 5.511111111111111,
|
| 559 |
+
"grad_norm": 1.2177822589874268,
|
| 560 |
+
"learning_rate": 2.4688342135114625e-06,
|
| 561 |
+
"logits/chosen": -0.18045730888843536,
|
| 562 |
+
"logits/rejected": -0.2632906138896942,
|
| 563 |
+
"logps/chosen": -8.068435668945312,
|
| 564 |
+
"logps/rejected": -38.28617477416992,
|
| 565 |
+
"loss": 0.2432,
|
| 566 |
+
"rewards/accuracies": 0.893750011920929,
|
| 567 |
+
"rewards/chosen": 1.5070723295211792,
|
| 568 |
+
"rewards/margins": 2.8888046741485596,
|
| 569 |
+
"rewards/rejected": -1.3817322254180908,
|
| 570 |
+
"step": 310
|
| 571 |
+
},
|
| 572 |
+
{
|
| 573 |
+
"epoch": 5.688888888888889,
|
| 574 |
+
"grad_norm": 6.162428379058838,
|
| 575 |
+
"learning_rate": 2.3131747660339396e-06,
|
| 576 |
+
"logits/chosen": -0.16258001327514648,
|
| 577 |
+
"logits/rejected": -0.2531837522983551,
|
| 578 |
+
"logps/chosen": -7.69020938873291,
|
| 579 |
+
"logps/rejected": -39.50102233886719,
|
| 580 |
+
"loss": 0.2155,
|
| 581 |
+
"rewards/accuracies": 0.893750011920929,
|
| 582 |
+
"rewards/chosen": 1.4402731657028198,
|
| 583 |
+
"rewards/margins": 2.9210917949676514,
|
| 584 |
+
"rewards/rejected": -1.4808186292648315,
|
| 585 |
+
"step": 320
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 5.866666666666667,
|
| 589 |
+
"grad_norm": 4.585377216339111,
|
| 590 |
+
"learning_rate": 2.158240979224817e-06,
|
| 591 |
+
"logits/chosen": -0.16741995513439178,
|
| 592 |
+
"logits/rejected": -0.25773563981056213,
|
| 593 |
+
"logps/chosen": -7.611827373504639,
|
| 594 |
+
"logps/rejected": -40.226985931396484,
|
| 595 |
+
"loss": 0.1765,
|
| 596 |
+
"rewards/accuracies": 0.918749988079071,
|
| 597 |
+
"rewards/chosen": 1.5339945554733276,
|
| 598 |
+
"rewards/margins": 3.106376886367798,
|
| 599 |
+
"rewards/rejected": -1.5723823308944702,
|
| 600 |
+
"step": 330
|
| 601 |
+
},
|
| 602 |
+
{
|
| 603 |
+
"epoch": 6.044444444444444,
|
| 604 |
+
"grad_norm": 1.262926697731018,
|
| 605 |
+
"learning_rate": 2.004634642001507e-06,
|
| 606 |
+
"logits/chosen": -0.17107848823070526,
|
| 607 |
+
"logits/rejected": -0.2691803276538849,
|
| 608 |
+
"logps/chosen": -8.973631858825684,
|
| 609 |
+
"logps/rejected": -41.54354476928711,
|
| 610 |
+
"loss": 0.1994,
|
| 611 |
+
"rewards/accuracies": 0.9312500357627869,
|
| 612 |
+
"rewards/chosen": 1.4238203763961792,
|
| 613 |
+
"rewards/margins": 3.09421968460083,
|
| 614 |
+
"rewards/rejected": -1.67039954662323,
|
| 615 |
+
"step": 340
|
| 616 |
+
},
|
| 617 |
+
{
|
| 618 |
+
"epoch": 6.222222222222222,
|
| 619 |
+
"grad_norm": 7.014649868011475,
|
| 620 |
+
"learning_rate": 1.852952387243698e-06,
|
| 621 |
+
"logits/chosen": -0.17273230850696564,
|
| 622 |
+
"logits/rejected": -0.2703326344490051,
|
| 623 |
+
"logps/chosen": -8.288504600524902,
|
| 624 |
+
"logps/rejected": -42.46126174926758,
|
| 625 |
+
"loss": 0.2047,
|
| 626 |
+
"rewards/accuracies": 0.918749988079071,
|
| 627 |
+
"rewards/chosen": 1.4250819683074951,
|
| 628 |
+
"rewards/margins": 3.235600709915161,
|
| 629 |
+
"rewards/rejected": -1.8105186223983765,
|
| 630 |
+
"step": 350
|
| 631 |
+
},
|
| 632 |
+
{
|
| 633 |
+
"epoch": 6.222222222222222,
|
| 634 |
+
"eval_logits/chosen": -0.15530475974082947,
|
| 635 |
+
"eval_logits/rejected": -0.2578139305114746,
|
| 636 |
+
"eval_logps/chosen": -9.47491455078125,
|
| 637 |
+
"eval_logps/rejected": -40.34950637817383,
|
| 638 |
+
"eval_loss": 0.3629891574382782,
|
| 639 |
+
"eval_rewards/accuracies": 0.8399999737739563,
|
| 640 |
+
"eval_rewards/chosen": 1.3018722534179688,
|
| 641 |
+
"eval_rewards/margins": 2.897495746612549,
|
| 642 |
+
"eval_rewards/rejected": -1.59562349319458,
|
| 643 |
+
"eval_runtime": 13.1284,
|
| 644 |
+
"eval_samples_per_second": 7.617,
|
| 645 |
+
"eval_steps_per_second": 3.809,
|
| 646 |
+
"step": 350
|
| 647 |
+
},
|
| 648 |
+
{
|
| 649 |
+
"epoch": 6.4,
|
| 650 |
+
"grad_norm": 8.984504699707031,
|
| 651 |
+
"learning_rate": 1.7037833743707892e-06,
|
| 652 |
+
"logits/chosen": -0.14409998059272766,
|
| 653 |
+
"logits/rejected": -0.2163417637348175,
|
| 654 |
+
"logps/chosen": -8.419466018676758,
|
| 655 |
+
"logps/rejected": -41.356468200683594,
|
| 656 |
+
"loss": 0.1894,
|
| 657 |
+
"rewards/accuracies": 0.9000000357627869,
|
| 658 |
+
"rewards/chosen": 1.4324077367782593,
|
| 659 |
+
"rewards/margins": 3.1765902042388916,
|
| 660 |
+
"rewards/rejected": -1.7441825866699219,
|
| 661 |
+
"step": 360
|
| 662 |
+
},
|
| 663 |
+
{
|
| 664 |
+
"epoch": 6.5777777777777775,
|
| 665 |
+
"grad_norm": 7.33497953414917,
|
| 666 |
+
"learning_rate": 1.5577070009474872e-06,
|
| 667 |
+
"logits/chosen": -0.14226722717285156,
|
| 668 |
+
"logits/rejected": -0.23484404385089874,
|
| 669 |
+
"logps/chosen": -7.855214595794678,
|
| 670 |
+
"logps/rejected": -42.96932601928711,
|
| 671 |
+
"loss": 0.19,
|
| 672 |
+
"rewards/accuracies": 0.9000000357627869,
|
| 673 |
+
"rewards/chosen": 1.4563724994659424,
|
| 674 |
+
"rewards/margins": 3.2921605110168457,
|
| 675 |
+
"rewards/rejected": -1.8357880115509033,
|
| 676 |
+
"step": 370
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 6.7555555555555555,
|
| 680 |
+
"grad_norm": 1.3846873044967651,
|
| 681 |
+
"learning_rate": 1.415290652206105e-06,
|
| 682 |
+
"logits/chosen": -0.11893842369318008,
|
| 683 |
+
"logits/rejected": -0.22303898632526398,
|
| 684 |
+
"logps/chosen": -5.515126705169678,
|
| 685 |
+
"logps/rejected": -43.147682189941406,
|
| 686 |
+
"loss": 0.1296,
|
| 687 |
+
"rewards/accuracies": 0.96875,
|
| 688 |
+
"rewards/chosen": 1.6900787353515625,
|
| 689 |
+
"rewards/margins": 3.544553756713867,
|
| 690 |
+
"rewards/rejected": -1.8544749021530151,
|
| 691 |
+
"step": 380
|
| 692 |
+
},
|
| 693 |
+
{
|
| 694 |
+
"epoch": 6.933333333333334,
|
| 695 |
+
"grad_norm": 3.4178779125213623,
|
| 696 |
+
"learning_rate": 1.2770874972267777e-06,
|
| 697 |
+
"logits/chosen": -0.2029663324356079,
|
| 698 |
+
"logits/rejected": -0.27416250109672546,
|
| 699 |
+
"logps/chosen": -10.447962760925293,
|
| 700 |
+
"logps/rejected": -41.886138916015625,
|
| 701 |
+
"loss": 0.2833,
|
| 702 |
+
"rewards/accuracies": 0.925000011920929,
|
| 703 |
+
"rewards/chosen": 1.2835975885391235,
|
| 704 |
+
"rewards/margins": 3.022332191467285,
|
| 705 |
+
"rewards/rejected": -1.738734483718872,
|
| 706 |
+
"step": 390
|
| 707 |
+
},
|
| 708 |
+
{
|
| 709 |
+
"epoch": 7.111111111111111,
|
| 710 |
+
"grad_norm": 5.459321975708008,
|
| 711 |
+
"learning_rate": 1.1436343403356019e-06,
|
| 712 |
+
"logits/chosen": -0.14208023250102997,
|
| 713 |
+
"logits/rejected": -0.22397640347480774,
|
| 714 |
+
"logps/chosen": -8.451988220214844,
|
| 715 |
+
"logps/rejected": -42.06145095825195,
|
| 716 |
+
"loss": 0.2268,
|
| 717 |
+
"rewards/accuracies": 0.893750011920929,
|
| 718 |
+
"rewards/chosen": 1.4957071542739868,
|
| 719 |
+
"rewards/margins": 3.238795518875122,
|
| 720 |
+
"rewards/rejected": -1.7430883646011353,
|
| 721 |
+
"step": 400
|
| 722 |
+
},
|
| 723 |
+
{
|
| 724 |
+
"epoch": 7.111111111111111,
|
| 725 |
+
"eval_logits/chosen": -0.1451883465051651,
|
| 726 |
+
"eval_logits/rejected": -0.24787528812885284,
|
| 727 |
+
"eval_logps/chosen": -8.884222030639648,
|
| 728 |
+
"eval_logps/rejected": -41.02870178222656,
|
| 729 |
+
"eval_loss": 0.352620393037796,
|
| 730 |
+
"eval_rewards/accuracies": 0.8499999642372131,
|
| 731 |
+
"eval_rewards/chosen": 1.3609414100646973,
|
| 732 |
+
"eval_rewards/margins": 3.024484872817993,
|
| 733 |
+
"eval_rewards/rejected": -1.6635433435440063,
|
| 734 |
+
"eval_runtime": 13.1853,
|
| 735 |
+
"eval_samples_per_second": 7.584,
|
| 736 |
+
"eval_steps_per_second": 3.792,
|
| 737 |
+
"step": 400
|
| 738 |
+
},
|
| 739 |
+
{
|
| 740 |
+
"epoch": 7.288888888888889,
|
| 741 |
+
"grad_norm": 2.762150764465332,
|
| 742 |
+
"learning_rate": 1.0154495360662464e-06,
|
| 743 |
+
"logits/chosen": -0.18289707601070404,
|
| 744 |
+
"logits/rejected": -0.28008756041526794,
|
| 745 |
+
"logps/chosen": -7.561173439025879,
|
| 746 |
+
"logps/rejected": -43.28544998168945,
|
| 747 |
+
"loss": 0.1976,
|
| 748 |
+
"rewards/accuracies": 0.925000011920929,
|
| 749 |
+
"rewards/chosen": 1.419569730758667,
|
| 750 |
+
"rewards/margins": 3.3263187408447266,
|
| 751 |
+
"rewards/rejected": -1.9067490100860596,
|
| 752 |
+
"step": 410
|
| 753 |
+
},
|
| 754 |
+
{
|
| 755 |
+
"epoch": 7.466666666666667,
|
| 756 |
+
"grad_norm": 2.009507417678833,
|
| 757 |
+
"learning_rate": 8.930309757836517e-07,
|
| 758 |
+
"logits/chosen": -0.12424879521131516,
|
| 759 |
+
"logits/rejected": -0.22191472351551056,
|
| 760 |
+
"logps/chosen": -7.338543891906738,
|
| 761 |
+
"logps/rejected": -42.68900680541992,
|
| 762 |
+
"loss": 0.2035,
|
| 763 |
+
"rewards/accuracies": 0.90625,
|
| 764 |
+
"rewards/chosen": 1.533159852027893,
|
| 765 |
+
"rewards/margins": 3.3633785247802734,
|
| 766 |
+
"rewards/rejected": -1.8302189111709595,
|
| 767 |
+
"step": 420
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 7.644444444444445,
|
| 771 |
+
"grad_norm": 1.1898467540740967,
|
| 772 |
+
"learning_rate": 7.768541537901325e-07,
|
| 773 |
+
"logits/chosen": -0.13503606617450714,
|
| 774 |
+
"logits/rejected": -0.23675648868083954,
|
| 775 |
+
"logps/chosen": -8.620238304138184,
|
| 776 |
+
"logps/rejected": -44.113094329833984,
|
| 777 |
+
"loss": 0.1728,
|
| 778 |
+
"rewards/accuracies": 0.918749988079071,
|
| 779 |
+
"rewards/chosen": 1.4350792169570923,
|
| 780 |
+
"rewards/margins": 3.3960673809051514,
|
| 781 |
+
"rewards/rejected": -1.9609882831573486,
|
| 782 |
+
"step": 430
|
| 783 |
+
},
|
| 784 |
+
{
|
| 785 |
+
"epoch": 7.822222222222222,
|
| 786 |
+
"grad_norm": 6.634049892425537,
|
| 787 |
+
"learning_rate": 6.673703204254348e-07,
|
| 788 |
+
"logits/chosen": -0.13498146831989288,
|
| 789 |
+
"logits/rejected": -0.21733498573303223,
|
| 790 |
+
"logps/chosen": -8.704492568969727,
|
| 791 |
+
"logps/rejected": -43.417938232421875,
|
| 792 |
+
"loss": 0.2495,
|
| 793 |
+
"rewards/accuracies": 0.918749988079071,
|
| 794 |
+
"rewards/chosen": 1.4058576822280884,
|
| 795 |
+
"rewards/margins": 3.286863088607788,
|
| 796 |
+
"rewards/rejected": -1.8810051679611206,
|
| 797 |
+
"step": 440
|
| 798 |
+
},
|
| 799 |
+
{
|
| 800 |
+
"epoch": 8.0,
|
| 801 |
+
"grad_norm": 4.482236385345459,
|
| 802 |
+
"learning_rate": 5.650047293344316e-07,
|
| 803 |
+
"logits/chosen": -0.14725255966186523,
|
| 804 |
+
"logits/rejected": -0.2412375509738922,
|
| 805 |
+
"logps/chosen": -7.016914367675781,
|
| 806 |
+
"logps/rejected": -44.169490814208984,
|
| 807 |
+
"loss": 0.144,
|
| 808 |
+
"rewards/accuracies": 0.9375,
|
| 809 |
+
"rewards/chosen": 1.5698460340499878,
|
| 810 |
+
"rewards/margins": 3.528562307357788,
|
| 811 |
+
"rewards/rejected": -1.9587162733078003,
|
| 812 |
+
"step": 450
|
| 813 |
+
},
|
| 814 |
+
{
|
| 815 |
+
"epoch": 8.0,
|
| 816 |
+
"eval_logits/chosen": -0.14205148816108704,
|
| 817 |
+
"eval_logits/rejected": -0.24482183158397675,
|
| 818 |
+
"eval_logps/chosen": -9.005942344665527,
|
| 819 |
+
"eval_logps/rejected": -41.425498962402344,
|
| 820 |
+
"eval_loss": 0.3662210702896118,
|
| 821 |
+
"eval_rewards/accuracies": 0.8399999737739563,
|
| 822 |
+
"eval_rewards/chosen": 1.3487695455551147,
|
| 823 |
+
"eval_rewards/margins": 3.051992654800415,
|
| 824 |
+
"eval_rewards/rejected": -1.7032231092453003,
|
| 825 |
+
"eval_runtime": 13.1525,
|
| 826 |
+
"eval_samples_per_second": 7.603,
|
| 827 |
+
"eval_steps_per_second": 3.802,
|
| 828 |
+
"step": 450
|
| 829 |
+
},
|
| 830 |
+
{
|
| 831 |
+
"epoch": 8.177777777777777,
|
| 832 |
+
"grad_norm": 18.307058334350586,
|
| 833 |
+
"learning_rate": 4.7015498571035877e-07,
|
| 834 |
+
"logits/chosen": -0.10620441287755966,
|
| 835 |
+
"logits/rejected": -0.19579415023326874,
|
| 836 |
+
"logps/chosen": -9.014161109924316,
|
| 837 |
+
"logps/rejected": -43.37262725830078,
|
| 838 |
+
"loss": 0.2079,
|
| 839 |
+
"rewards/accuracies": 0.918749988079071,
|
| 840 |
+
"rewards/chosen": 1.3402118682861328,
|
| 841 |
+
"rewards/margins": 3.250830888748169,
|
| 842 |
+
"rewards/rejected": -1.910618782043457,
|
| 843 |
+
"step": 460
|
| 844 |
+
},
|
| 845 |
+
{
|
| 846 |
+
"epoch": 8.355555555555556,
|
| 847 |
+
"grad_norm": 1.149519681930542,
|
| 848 |
+
"learning_rate": 3.831895019292897e-07,
|
| 849 |
+
"logits/chosen": -0.12923561036586761,
|
| 850 |
+
"logits/rejected": -0.24200458824634552,
|
| 851 |
+
"logps/chosen": -6.25982666015625,
|
| 852 |
+
"logps/rejected": -44.513946533203125,
|
| 853 |
+
"loss": 0.1659,
|
| 854 |
+
"rewards/accuracies": 0.9125000238418579,
|
| 855 |
+
"rewards/chosen": 1.5601545572280884,
|
| 856 |
+
"rewards/margins": 3.614009141921997,
|
| 857 |
+
"rewards/rejected": -2.0538547039031982,
|
| 858 |
+
"step": 470
|
| 859 |
+
},
|
| 860 |
+
{
|
| 861 |
+
"epoch": 8.533333333333333,
|
| 862 |
+
"grad_norm": 9.660409927368164,
|
| 863 |
+
"learning_rate": 3.044460665744284e-07,
|
| 864 |
+
"logits/chosen": -0.12687157094478607,
|
| 865 |
+
"logits/rejected": -0.2105720490217209,
|
| 866 |
+
"logps/chosen": -8.170889854431152,
|
| 867 |
+
"logps/rejected": -43.91897201538086,
|
| 868 |
+
"loss": 0.2133,
|
| 869 |
+
"rewards/accuracies": 0.9000000357627869,
|
| 870 |
+
"rewards/chosen": 1.490912675857544,
|
| 871 |
+
"rewards/margins": 3.416290283203125,
|
| 872 |
+
"rewards/rejected": -1.9253777265548706,
|
| 873 |
+
"step": 480
|
| 874 |
+
},
|
| 875 |
+
{
|
| 876 |
+
"epoch": 8.71111111111111,
|
| 877 |
+
"grad_norm": 6.015744686126709,
|
| 878 |
+
"learning_rate": 2.3423053240837518e-07,
|
| 879 |
+
"logits/chosen": -0.1734294891357422,
|
| 880 |
+
"logits/rejected": -0.2597627341747284,
|
| 881 |
+
"logps/chosen": -9.138812065124512,
|
| 882 |
+
"logps/rejected": -43.02109909057617,
|
| 883 |
+
"loss": 0.1926,
|
| 884 |
+
"rewards/accuracies": 0.925000011920929,
|
| 885 |
+
"rewards/chosen": 1.4286350011825562,
|
| 886 |
+
"rewards/margins": 3.2899787425994873,
|
| 887 |
+
"rewards/rejected": -1.8613442182540894,
|
| 888 |
+
"step": 490
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 8.88888888888889,
|
| 892 |
+
"grad_norm": 5.968247890472412,
|
| 893 |
+
"learning_rate": 1.7281562838948968e-07,
|
| 894 |
+
"logits/chosen": -0.1330401748418808,
|
| 895 |
+
"logits/rejected": -0.2250121682882309,
|
| 896 |
+
"logps/chosen": -7.425269603729248,
|
| 897 |
+
"logps/rejected": -43.31748580932617,
|
| 898 |
+
"loss": 0.171,
|
| 899 |
+
"rewards/accuracies": 0.956250011920929,
|
| 900 |
+
"rewards/chosen": 1.5740032196044922,
|
| 901 |
+
"rewards/margins": 3.423919677734375,
|
| 902 |
+
"rewards/rejected": -1.8499164581298828,
|
| 903 |
+
"step": 500
|
| 904 |
+
},
|
| 905 |
+
{
|
| 906 |
+
"epoch": 8.88888888888889,
|
| 907 |
+
"eval_logits/chosen": -0.1399160772562027,
|
| 908 |
+
"eval_logits/rejected": -0.2429780215024948,
|
| 909 |
+
"eval_logps/chosen": -9.18053913116455,
|
| 910 |
+
"eval_logps/rejected": -41.719730377197266,
|
| 911 |
+
"eval_loss": 0.36349406838417053,
|
| 912 |
+
"eval_rewards/accuracies": 0.8399999737739563,
|
| 913 |
+
"eval_rewards/chosen": 1.3313097953796387,
|
| 914 |
+
"eval_rewards/margins": 3.0639562606811523,
|
| 915 |
+
"eval_rewards/rejected": -1.7326463460922241,
|
| 916 |
+
"eval_runtime": 13.2218,
|
| 917 |
+
"eval_samples_per_second": 7.563,
|
| 918 |
+
"eval_steps_per_second": 3.782,
|
| 919 |
+
"step": 500
|
| 920 |
+
},
|
| 921 |
+
{
|
| 922 |
+
"epoch": 9.066666666666666,
|
| 923 |
+
"grad_norm": 2.3828182220458984,
|
| 924 |
+
"learning_rate": 1.2043990034669413e-07,
|
| 925 |
+
"logits/chosen": -0.16531167924404144,
|
| 926 |
+
"logits/rejected": -0.26316291093826294,
|
| 927 |
+
"logps/chosen": -6.18184757232666,
|
| 928 |
+
"logps/rejected": -45.66379928588867,
|
| 929 |
+
"loss": 0.136,
|
| 930 |
+
"rewards/accuracies": 0.949999988079071,
|
| 931 |
+
"rewards/chosen": 1.5860216617584229,
|
| 932 |
+
"rewards/margins": 3.676870107650757,
|
| 933 |
+
"rewards/rejected": -2.090848207473755,
|
| 934 |
+
"step": 510
|
| 935 |
+
},
|
| 936 |
+
{
|
| 937 |
+
"epoch": 9.244444444444444,
|
| 938 |
+
"grad_norm": 1.0561579465866089,
|
| 939 |
+
"learning_rate": 7.730678442730539e-08,
|
| 940 |
+
"logits/chosen": -0.12930361926555634,
|
| 941 |
+
"logits/rejected": -0.22025151550769806,
|
| 942 |
+
"logps/chosen": -7.336406230926514,
|
| 943 |
+
"logps/rejected": -44.349239349365234,
|
| 944 |
+
"loss": 0.1573,
|
| 945 |
+
"rewards/accuracies": 0.9375,
|
| 946 |
+
"rewards/chosen": 1.5953136682510376,
|
| 947 |
+
"rewards/margins": 3.572978973388672,
|
| 948 |
+
"rewards/rejected": -1.9776651859283447,
|
| 949 |
+
"step": 520
|
| 950 |
+
},
|
| 951 |
+
{
|
| 952 |
+
"epoch": 9.422222222222222,
|
| 953 |
+
"grad_norm": 11.791927337646484,
|
| 954 |
+
"learning_rate": 4.358381691677932e-08,
|
| 955 |
+
"logits/chosen": -0.10277407616376877,
|
| 956 |
+
"logits/rejected": -0.1993442177772522,
|
| 957 |
+
"logps/chosen": -7.81331205368042,
|
| 958 |
+
"logps/rejected": -44.955955505371094,
|
| 959 |
+
"loss": 0.1684,
|
| 960 |
+
"rewards/accuracies": 0.925000011920929,
|
| 961 |
+
"rewards/chosen": 1.5365545749664307,
|
| 962 |
+
"rewards/margins": 3.621255874633789,
|
| 963 |
+
"rewards/rejected": -2.0847015380859375,
|
| 964 |
+
"step": 530
|
| 965 |
+
},
|
| 966 |
+
{
|
| 967 |
+
"epoch": 9.6,
|
| 968 |
+
"grad_norm": 3.8846185207366943,
|
| 969 |
+
"learning_rate": 1.9401983499569843e-08,
|
| 970 |
+
"logits/chosen": -0.1286163181066513,
|
| 971 |
+
"logits/rejected": -0.25202152132987976,
|
| 972 |
+
"logps/chosen": -7.090148448944092,
|
| 973 |
+
"logps/rejected": -45.3367805480957,
|
| 974 |
+
"loss": 0.1675,
|
| 975 |
+
"rewards/accuracies": 0.925000011920929,
|
| 976 |
+
"rewards/chosen": 1.5582847595214844,
|
| 977 |
+
"rewards/margins": 3.6432290077209473,
|
| 978 |
+
"rewards/rejected": -2.084944009780884,
|
| 979 |
+
"step": 540
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 9.777777777777779,
|
| 983 |
+
"grad_norm": 6.826923847198486,
|
| 984 |
+
"learning_rate": 4.855210488670381e-09,
|
| 985 |
+
"logits/chosen": -0.14255917072296143,
|
| 986 |
+
"logits/rejected": -0.24119222164154053,
|
| 987 |
+
"logps/chosen": -9.036189079284668,
|
| 988 |
+
"logps/rejected": -41.72261047363281,
|
| 989 |
+
"loss": 0.2313,
|
| 990 |
+
"rewards/accuracies": 0.9125000238418579,
|
| 991 |
+
"rewards/chosen": 1.3214199542999268,
|
| 992 |
+
"rewards/margins": 3.085725784301758,
|
| 993 |
+
"rewards/rejected": -1.7643059492111206,
|
| 994 |
+
"step": 550
|
| 995 |
+
},
|
| 996 |
+
{
|
| 997 |
+
"epoch": 9.777777777777779,
|
| 998 |
+
"eval_logits/chosen": -0.13784560561180115,
|
| 999 |
+
"eval_logits/rejected": -0.24095743894577026,
|
| 1000 |
+
"eval_logps/chosen": -9.1017484664917,
|
| 1001 |
+
"eval_logps/rejected": -41.82564926147461,
|
| 1002 |
+
"eval_loss": 0.3612578511238098,
|
| 1003 |
+
"eval_rewards/accuracies": 0.8399999737739563,
|
| 1004 |
+
"eval_rewards/chosen": 1.3391889333724976,
|
| 1005 |
+
"eval_rewards/margins": 3.0824267864227295,
|
| 1006 |
+
"eval_rewards/rejected": -1.7432377338409424,
|
| 1007 |
+
"eval_runtime": 13.1911,
|
| 1008 |
+
"eval_samples_per_second": 7.581,
|
| 1009 |
+
"eval_steps_per_second": 3.79,
|
| 1010 |
+
"step": 550
|
| 1011 |
+
},
|
| 1012 |
+
{
|
| 1013 |
+
"epoch": 9.955555555555556,
|
| 1014 |
+
"grad_norm": 1.7594248056411743,
|
| 1015 |
+
"learning_rate": 0.0,
|
| 1016 |
+
"logits/chosen": -0.15151795744895935,
|
| 1017 |
+
"logits/rejected": -0.2365965098142624,
|
| 1018 |
+
"logps/chosen": -7.8617448806762695,
|
| 1019 |
+
"logps/rejected": -42.87919235229492,
|
| 1020 |
+
"loss": 0.1746,
|
| 1021 |
+
"rewards/accuracies": 0.9312500357627869,
|
| 1022 |
+
"rewards/chosen": 1.5052164793014526,
|
| 1023 |
+
"rewards/margins": 3.3184120655059814,
|
| 1024 |
+
"rewards/rejected": -1.8131954669952393,
|
| 1025 |
+
"step": 560
|
| 1026 |
+
},
|
| 1027 |
+
{
|
| 1028 |
+
"epoch": 9.955555555555556,
|
| 1029 |
+
"step": 560,
|
| 1030 |
+
"total_flos": 6.741083695664333e+16,
|
| 1031 |
+
"train_loss": 0.31093634622437616,
|
| 1032 |
+
"train_runtime": 2833.3873,
|
| 1033 |
+
"train_samples_per_second": 3.176,
|
| 1034 |
+
"train_steps_per_second": 0.198
|
| 1035 |
+
}
|
| 1036 |
+
],
|
| 1037 |
+
"logging_steps": 10,
|
| 1038 |
+
"max_steps": 560,
|
| 1039 |
+
"num_input_tokens_seen": 0,
|
| 1040 |
+
"num_train_epochs": 10,
|
| 1041 |
+
"save_steps": 50,
|
| 1042 |
+
"stateful_callbacks": {
|
| 1043 |
+
"TrainerControl": {
|
| 1044 |
+
"args": {
|
| 1045 |
+
"should_epoch_stop": false,
|
| 1046 |
+
"should_evaluate": false,
|
| 1047 |
+
"should_log": false,
|
| 1048 |
+
"should_save": true,
|
| 1049 |
+
"should_training_stop": true
|
| 1050 |
+
},
|
| 1051 |
+
"attributes": {}
|
| 1052 |
+
}
|
| 1053 |
+
},
|
| 1054 |
+
"total_flos": 6.741083695664333e+16,
|
| 1055 |
+
"train_batch_size": 2,
|
| 1056 |
+
"trial_name": null,
|
| 1057 |
+
"trial_params": null
|
| 1058 |
+
}
|
training_eval_loss.png
ADDED
|
training_loss.png
ADDED
|
training_rewards_accuracies.png
ADDED
|