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-mistral-1000
|
|
@@ -17,17 +18,17 @@ should probably proofread and complete it, then remove this comment. -->
|
|
| 17 |
|
| 18 |
# Llama-3.1-8B-Instruct-dpo-mistral-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: 0.
|
| 24 |
-
- Rewards/rejected: -0.
|
| 25 |
-
- Rewards/accuracies: 0.
|
| 26 |
-
- Rewards/margins: 1.
|
| 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-mistral-1000
|
|
|
|
| 18 |
|
| 19 |
# Llama-3.1-8B-Instruct-dpo-mistral-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_mistral dataset.
|
| 22 |
It achieves the following results on the evaluation set:
|
| 23 |
+
- Loss: 0.4675
|
| 24 |
+
- Rewards/chosen: 0.9903
|
| 25 |
+
- Rewards/rejected: -0.3997
|
| 26 |
+
- Rewards/accuracies: 0.7900
|
| 27 |
+
- Rewards/margins: 1.3900
|
| 28 |
+
- Logps/chosen: -13.2488
|
| 29 |
+
- Logps/rejected: -29.2269
|
| 30 |
+
- Logits/chosen: -0.1396
|
| 31 |
+
- Logits/rejected: -0.2080
|
| 32 |
|
| 33 |
## Model description
|
| 34 |
|
all_results.json
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 9.977728285077951,
|
| 3 |
+
"eval_logits/chosen": -0.13963328301906586,
|
| 4 |
+
"eval_logits/rejected": -0.2079545557498932,
|
| 5 |
+
"eval_logps/chosen": -13.248819351196289,
|
| 6 |
+
"eval_logps/rejected": -29.226905822753906,
|
| 7 |
+
"eval_loss": 0.4674791693687439,
|
| 8 |
+
"eval_rewards/accuracies": 0.7899999618530273,
|
| 9 |
+
"eval_rewards/chosen": 0.9902573823928833,
|
| 10 |
+
"eval_rewards/margins": 1.390006184577942,
|
| 11 |
+
"eval_rewards/rejected": -0.3997488021850586,
|
| 12 |
+
"eval_runtime": 9.6246,
|
| 13 |
+
"eval_samples_per_second": 10.39,
|
| 14 |
+
"eval_steps_per_second": 5.195,
|
| 15 |
+
"total_flos": 5.23282185018409e+16,
|
| 16 |
+
"train_loss": 0.4115582968507494,
|
| 17 |
+
"train_runtime": 2017.6604,
|
| 18 |
+
"train_samples_per_second": 4.451,
|
| 19 |
+
"train_steps_per_second": 0.278
|
| 20 |
+
}
|
eval_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 9.977728285077951,
|
| 3 |
+
"eval_logits/chosen": -0.13963328301906586,
|
| 4 |
+
"eval_logits/rejected": -0.2079545557498932,
|
| 5 |
+
"eval_logps/chosen": -13.248819351196289,
|
| 6 |
+
"eval_logps/rejected": -29.226905822753906,
|
| 7 |
+
"eval_loss": 0.4674791693687439,
|
| 8 |
+
"eval_rewards/accuracies": 0.7899999618530273,
|
| 9 |
+
"eval_rewards/chosen": 0.9902573823928833,
|
| 10 |
+
"eval_rewards/margins": 1.390006184577942,
|
| 11 |
+
"eval_rewards/rejected": -0.3997488021850586,
|
| 12 |
+
"eval_runtime": 9.6246,
|
| 13 |
+
"eval_samples_per_second": 10.39,
|
| 14 |
+
"eval_steps_per_second": 5.195
|
| 15 |
+
}
|
train_results.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 9.977728285077951,
|
| 3 |
+
"total_flos": 5.23282185018409e+16,
|
| 4 |
+
"train_loss": 0.4115582968507494,
|
| 5 |
+
"train_runtime": 2017.6604,
|
| 6 |
+
"train_samples_per_second": 4.451,
|
| 7 |
+
"train_steps_per_second": 0.278
|
| 8 |
+
}
|
trainer_state.json
ADDED
|
@@ -0,0 +1,1058 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": 0.4674791693687439,
|
| 3 |
+
"best_model_checkpoint": "saves/sycophancy/Llama-3.1-8B-Instruct/dpo-mistral-1000/train/checkpoint-250",
|
| 4 |
+
"epoch": 9.977728285077951,
|
| 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": 0.17817371937639198,
|
| 13 |
+
"grad_norm": 1.6836061477661133,
|
| 14 |
+
"learning_rate": 8.928571428571429e-07,
|
| 15 |
+
"logits/chosen": -0.34122365713119507,
|
| 16 |
+
"logits/rejected": -0.392149418592453,
|
| 17 |
+
"logps/chosen": -22.754064559936523,
|
| 18 |
+
"logps/rejected": -24.806787490844727,
|
| 19 |
+
"loss": 0.6925,
|
| 20 |
+
"rewards/accuracies": 0.4000000059604645,
|
| 21 |
+
"rewards/chosen": -0.00038957837386988103,
|
| 22 |
+
"rewards/margins": 0.0013258380349725485,
|
| 23 |
+
"rewards/rejected": -0.0017154163215309381,
|
| 24 |
+
"step": 10
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.35634743875278396,
|
| 28 |
+
"grad_norm": 1.5660648345947266,
|
| 29 |
+
"learning_rate": 1.7857142857142859e-06,
|
| 30 |
+
"logits/chosen": -0.3305678367614746,
|
| 31 |
+
"logits/rejected": -0.36542031168937683,
|
| 32 |
+
"logps/chosen": -22.68667221069336,
|
| 33 |
+
"logps/rejected": -24.383272171020508,
|
| 34 |
+
"loss": 0.694,
|
| 35 |
+
"rewards/accuracies": 0.46875,
|
| 36 |
+
"rewards/chosen": -0.0019774259999394417,
|
| 37 |
+
"rewards/margins": -0.0016633094055578113,
|
| 38 |
+
"rewards/rejected": -0.0003141165361739695,
|
| 39 |
+
"step": 20
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.534521158129176,
|
| 43 |
+
"grad_norm": 1.7988139390945435,
|
| 44 |
+
"learning_rate": 2.6785714285714285e-06,
|
| 45 |
+
"logits/chosen": -0.33457204699516296,
|
| 46 |
+
"logits/rejected": -0.3815780282020569,
|
| 47 |
+
"logps/chosen": -23.1612548828125,
|
| 48 |
+
"logps/rejected": -24.836502075195312,
|
| 49 |
+
"loss": 0.6913,
|
| 50 |
+
"rewards/accuracies": 0.6000000238418579,
|
| 51 |
+
"rewards/chosen": 0.004869468975812197,
|
| 52 |
+
"rewards/margins": 0.0037120466586202383,
|
| 53 |
+
"rewards/rejected": 0.001157421967945993,
|
| 54 |
+
"step": 30
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"epoch": 0.7126948775055679,
|
| 58 |
+
"grad_norm": 1.3268821239471436,
|
| 59 |
+
"learning_rate": 3.5714285714285718e-06,
|
| 60 |
+
"logits/chosen": -0.3601827919483185,
|
| 61 |
+
"logits/rejected": -0.39572954177856445,
|
| 62 |
+
"logps/chosen": -22.90369987487793,
|
| 63 |
+
"logps/rejected": -24.570892333984375,
|
| 64 |
+
"loss": 0.6885,
|
| 65 |
+
"rewards/accuracies": 0.6312500238418579,
|
| 66 |
+
"rewards/chosen": 0.01631110906600952,
|
| 67 |
+
"rewards/margins": 0.009484974667429924,
|
| 68 |
+
"rewards/rejected": 0.006826136261224747,
|
| 69 |
+
"step": 40
|
| 70 |
+
},
|
| 71 |
+
{
|
| 72 |
+
"epoch": 0.89086859688196,
|
| 73 |
+
"grad_norm": 1.776597261428833,
|
| 74 |
+
"learning_rate": 4.464285714285715e-06,
|
| 75 |
+
"logits/chosen": -0.3659666180610657,
|
| 76 |
+
"logits/rejected": -0.39071187376976013,
|
| 77 |
+
"logps/chosen": -23.908523559570312,
|
| 78 |
+
"logps/rejected": -24.262924194335938,
|
| 79 |
+
"loss": 0.6891,
|
| 80 |
+
"rewards/accuracies": 0.574999988079071,
|
| 81 |
+
"rewards/chosen": 0.026137981563806534,
|
| 82 |
+
"rewards/margins": 0.008756262250244617,
|
| 83 |
+
"rewards/rejected": 0.017381716519594193,
|
| 84 |
+
"step": 50
|
| 85 |
+
},
|
| 86 |
+
{
|
| 87 |
+
"epoch": 0.89086859688196,
|
| 88 |
+
"eval_logits/chosen": -0.32067814469337463,
|
| 89 |
+
"eval_logits/rejected": -0.36896541714668274,
|
| 90 |
+
"eval_logps/chosen": -22.66472053527832,
|
| 91 |
+
"eval_logps/rejected": -24.953529357910156,
|
| 92 |
+
"eval_loss": 0.6832554340362549,
|
| 93 |
+
"eval_rewards/accuracies": 0.6200000047683716,
|
| 94 |
+
"eval_rewards/chosen": 0.048666905611753464,
|
| 95 |
+
"eval_rewards/margins": 0.02107813023030758,
|
| 96 |
+
"eval_rewards/rejected": 0.027588771656155586,
|
| 97 |
+
"eval_runtime": 9.7251,
|
| 98 |
+
"eval_samples_per_second": 10.283,
|
| 99 |
+
"eval_steps_per_second": 5.141,
|
| 100 |
+
"step": 50
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 1.069042316258352,
|
| 104 |
+
"grad_norm": 1.7856236696243286,
|
| 105 |
+
"learning_rate": 4.999222955002041e-06,
|
| 106 |
+
"logits/chosen": -0.35137540102005005,
|
| 107 |
+
"logits/rejected": -0.3762962818145752,
|
| 108 |
+
"logps/chosen": -22.667428970336914,
|
| 109 |
+
"logps/rejected": -24.934234619140625,
|
| 110 |
+
"loss": 0.675,
|
| 111 |
+
"rewards/accuracies": 0.612500011920929,
|
| 112 |
+
"rewards/chosen": 0.06478282064199448,
|
| 113 |
+
"rewards/margins": 0.0387502983212471,
|
| 114 |
+
"rewards/rejected": 0.026032526046037674,
|
| 115 |
+
"step": 60
|
| 116 |
+
},
|
| 117 |
+
{
|
| 118 |
+
"epoch": 1.247216035634744,
|
| 119 |
+
"grad_norm": 2.1098411083221436,
|
| 120 |
+
"learning_rate": 4.990486745229364e-06,
|
| 121 |
+
"logits/chosen": -0.34582623839378357,
|
| 122 |
+
"logits/rejected": -0.3823074400424957,
|
| 123 |
+
"logps/chosen": -21.945472717285156,
|
| 124 |
+
"logps/rejected": -23.575443267822266,
|
| 125 |
+
"loss": 0.6683,
|
| 126 |
+
"rewards/accuracies": 0.6500000357627869,
|
| 127 |
+
"rewards/chosen": 0.1263795644044876,
|
| 128 |
+
"rewards/margins": 0.05839619040489197,
|
| 129 |
+
"rewards/rejected": 0.06798337399959564,
|
| 130 |
+
"step": 70
|
| 131 |
+
},
|
| 132 |
+
{
|
| 133 |
+
"epoch": 1.4253897550111359,
|
| 134 |
+
"grad_norm": 2.4574801921844482,
|
| 135 |
+
"learning_rate": 4.9720770655628216e-06,
|
| 136 |
+
"logits/chosen": -0.3288664221763611,
|
| 137 |
+
"logits/rejected": -0.3651907444000244,
|
| 138 |
+
"logps/chosen": -21.0363712310791,
|
| 139 |
+
"logps/rejected": -24.36761474609375,
|
| 140 |
+
"loss": 0.6223,
|
| 141 |
+
"rewards/accuracies": 0.737500011920929,
|
| 142 |
+
"rewards/chosen": 0.26030755043029785,
|
| 143 |
+
"rewards/margins": 0.1713070124387741,
|
| 144 |
+
"rewards/rejected": 0.08900053799152374,
|
| 145 |
+
"step": 80
|
| 146 |
+
},
|
| 147 |
+
{
|
| 148 |
+
"epoch": 1.6035634743875278,
|
| 149 |
+
"grad_norm": 2.382906675338745,
|
| 150 |
+
"learning_rate": 4.944065422298262e-06,
|
| 151 |
+
"logits/chosen": -0.3467506468296051,
|
| 152 |
+
"logits/rejected": -0.38536104559898376,
|
| 153 |
+
"logps/chosen": -19.177209854125977,
|
| 154 |
+
"logps/rejected": -23.13625144958496,
|
| 155 |
+
"loss": 0.6035,
|
| 156 |
+
"rewards/accuracies": 0.699999988079071,
|
| 157 |
+
"rewards/chosen": 0.41219550371170044,
|
| 158 |
+
"rewards/margins": 0.25021034479141235,
|
| 159 |
+
"rewards/rejected": 0.1619851440191269,
|
| 160 |
+
"step": 90
|
| 161 |
+
},
|
| 162 |
+
{
|
| 163 |
+
"epoch": 1.7817371937639197,
|
| 164 |
+
"grad_norm": 2.123657703399658,
|
| 165 |
+
"learning_rate": 4.90656061737503e-06,
|
| 166 |
+
"logits/chosen": -0.3112158477306366,
|
| 167 |
+
"logits/rejected": -0.355999618768692,
|
| 168 |
+
"logps/chosen": -17.31022834777832,
|
| 169 |
+
"logps/rejected": -22.826955795288086,
|
| 170 |
+
"loss": 0.5716,
|
| 171 |
+
"rewards/accuracies": 0.699999988079071,
|
| 172 |
+
"rewards/chosen": 0.5237475633621216,
|
| 173 |
+
"rewards/margins": 0.35302406549453735,
|
| 174 |
+
"rewards/rejected": 0.17072349786758423,
|
| 175 |
+
"step": 100
|
| 176 |
+
},
|
| 177 |
+
{
|
| 178 |
+
"epoch": 1.7817371937639197,
|
| 179 |
+
"eval_logits/chosen": -0.2933846712112427,
|
| 180 |
+
"eval_logits/rejected": -0.34558260440826416,
|
| 181 |
+
"eval_logps/chosen": -17.070594787597656,
|
| 182 |
+
"eval_logps/rejected": -23.3165225982666,
|
| 183 |
+
"eval_loss": 0.5618187785148621,
|
| 184 |
+
"eval_rewards/accuracies": 0.699999988079071,
|
| 185 |
+
"eval_rewards/chosen": 0.6080796718597412,
|
| 186 |
+
"eval_rewards/margins": 0.4167901575565338,
|
| 187 |
+
"eval_rewards/rejected": 0.191289484500885,
|
| 188 |
+
"eval_runtime": 9.6952,
|
| 189 |
+
"eval_samples_per_second": 10.314,
|
| 190 |
+
"eval_steps_per_second": 5.157,
|
| 191 |
+
"step": 100
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 1.9599109131403119,
|
| 195 |
+
"grad_norm": 3.316772699356079,
|
| 196 |
+
"learning_rate": 4.859708325770919e-06,
|
| 197 |
+
"logits/chosen": -0.3041900396347046,
|
| 198 |
+
"logits/rejected": -0.35448360443115234,
|
| 199 |
+
"logps/chosen": -16.122028350830078,
|
| 200 |
+
"logps/rejected": -22.659238815307617,
|
| 201 |
+
"loss": 0.5299,
|
| 202 |
+
"rewards/accuracies": 0.7250000238418579,
|
| 203 |
+
"rewards/chosen": 0.7380782961845398,
|
| 204 |
+
"rewards/margins": 0.5233559608459473,
|
| 205 |
+
"rewards/rejected": 0.21472235023975372,
|
| 206 |
+
"step": 110
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"epoch": 2.138084632516704,
|
| 210 |
+
"grad_norm": 2.7125298976898193,
|
| 211 |
+
"learning_rate": 4.80369052967602e-06,
|
| 212 |
+
"logits/chosen": -0.2812982499599457,
|
| 213 |
+
"logits/rejected": -0.3268232047557831,
|
| 214 |
+
"logps/chosen": -15.398541450500488,
|
| 215 |
+
"logps/rejected": -23.535791397094727,
|
| 216 |
+
"loss": 0.4948,
|
| 217 |
+
"rewards/accuracies": 0.7750000357627869,
|
| 218 |
+
"rewards/chosen": 0.8048359155654907,
|
| 219 |
+
"rewards/margins": 0.660244882106781,
|
| 220 |
+
"rewards/rejected": 0.14459095895290375,
|
| 221 |
+
"step": 120
|
| 222 |
+
},
|
| 223 |
+
{
|
| 224 |
+
"epoch": 2.316258351893096,
|
| 225 |
+
"grad_norm": 2.149165391921997,
|
| 226 |
+
"learning_rate": 4.7387248116432524e-06,
|
| 227 |
+
"logits/chosen": -0.24634817242622375,
|
| 228 |
+
"logits/rejected": -0.313901424407959,
|
| 229 |
+
"logps/chosen": -13.057968139648438,
|
| 230 |
+
"logps/rejected": -23.697710037231445,
|
| 231 |
+
"loss": 0.4275,
|
| 232 |
+
"rewards/accuracies": 0.7750000357627869,
|
| 233 |
+
"rewards/chosen": 1.0445865392684937,
|
| 234 |
+
"rewards/margins": 0.944391667842865,
|
| 235 |
+
"rewards/rejected": 0.10019483417272568,
|
| 236 |
+
"step": 130
|
| 237 |
+
},
|
| 238 |
+
{
|
| 239 |
+
"epoch": 2.494432071269488,
|
| 240 |
+
"grad_norm": 2.2673017978668213,
|
| 241 |
+
"learning_rate": 4.665063509461098e-06,
|
| 242 |
+
"logits/chosen": -0.2440420240163803,
|
| 243 |
+
"logits/rejected": -0.29979586601257324,
|
| 244 |
+
"logps/chosen": -13.353204727172852,
|
| 245 |
+
"logps/rejected": -24.62551498413086,
|
| 246 |
+
"loss": 0.4575,
|
| 247 |
+
"rewards/accuracies": 0.7875000238418579,
|
| 248 |
+
"rewards/chosen": 0.9695870280265808,
|
| 249 |
+
"rewards/margins": 0.9478777050971985,
|
| 250 |
+
"rewards/rejected": 0.021709401160478592,
|
| 251 |
+
"step": 140
|
| 252 |
+
},
|
| 253 |
+
{
|
| 254 |
+
"epoch": 2.6726057906458798,
|
| 255 |
+
"grad_norm": 2.5609817504882812,
|
| 256 |
+
"learning_rate": 4.5829927360311224e-06,
|
| 257 |
+
"logits/chosen": -0.21621806919574738,
|
| 258 |
+
"logits/rejected": -0.27183279395103455,
|
| 259 |
+
"logps/chosen": -12.837681770324707,
|
| 260 |
+
"logps/rejected": -24.354141235351562,
|
| 261 |
+
"loss": 0.4581,
|
| 262 |
+
"rewards/accuracies": 0.7750000357627869,
|
| 263 |
+
"rewards/chosen": 0.9683197140693665,
|
| 264 |
+
"rewards/margins": 0.9282311797142029,
|
| 265 |
+
"rewards/rejected": 0.04008860886096954,
|
| 266 |
+
"step": 150
|
| 267 |
+
},
|
| 268 |
+
{
|
| 269 |
+
"epoch": 2.6726057906458798,
|
| 270 |
+
"eval_logits/chosen": -0.2092563956975937,
|
| 271 |
+
"eval_logits/rejected": -0.2738884389400482,
|
| 272 |
+
"eval_logps/chosen": -13.789220809936523,
|
| 273 |
+
"eval_logps/rejected": -25.666584014892578,
|
| 274 |
+
"eval_loss": 0.4760795533657074,
|
| 275 |
+
"eval_rewards/accuracies": 0.7599999904632568,
|
| 276 |
+
"eval_rewards/chosen": 0.9362172484397888,
|
| 277 |
+
"eval_rewards/margins": 0.9799338579177856,
|
| 278 |
+
"eval_rewards/rejected": -0.04371662810444832,
|
| 279 |
+
"eval_runtime": 9.724,
|
| 280 |
+
"eval_samples_per_second": 10.284,
|
| 281 |
+
"eval_steps_per_second": 5.142,
|
| 282 |
+
"step": 150
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 2.8507795100222717,
|
| 286 |
+
"grad_norm": 6.471484661102295,
|
| 287 |
+
"learning_rate": 4.492831268057307e-06,
|
| 288 |
+
"logits/chosen": -0.25587916374206543,
|
| 289 |
+
"logits/rejected": -0.3184296786785126,
|
| 290 |
+
"logps/chosen": -12.521145820617676,
|
| 291 |
+
"logps/rejected": -25.58576011657715,
|
| 292 |
+
"loss": 0.406,
|
| 293 |
+
"rewards/accuracies": 0.7750000357627869,
|
| 294 |
+
"rewards/chosen": 1.077256202697754,
|
| 295 |
+
"rewards/margins": 1.1716722249984741,
|
| 296 |
+
"rewards/rejected": -0.09441610425710678,
|
| 297 |
+
"step": 160
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"epoch": 3.0289532293986636,
|
| 301 |
+
"grad_norm": 6.336172580718994,
|
| 302 |
+
"learning_rate": 4.394929307863633e-06,
|
| 303 |
+
"logits/chosen": -0.23687370121479034,
|
| 304 |
+
"logits/rejected": -0.2813915014266968,
|
| 305 |
+
"logps/chosen": -14.616838455200195,
|
| 306 |
+
"logps/rejected": -26.1534366607666,
|
| 307 |
+
"loss": 0.452,
|
| 308 |
+
"rewards/accuracies": 0.800000011920929,
|
| 309 |
+
"rewards/chosen": 0.8590637445449829,
|
| 310 |
+
"rewards/margins": 1.0506114959716797,
|
| 311 |
+
"rewards/rejected": -0.19154782593250275,
|
| 312 |
+
"step": 170
|
| 313 |
+
},
|
| 314 |
+
{
|
| 315 |
+
"epoch": 3.2071269487750556,
|
| 316 |
+
"grad_norm": 11.07475471496582,
|
| 317 |
+
"learning_rate": 4.289667123149296e-06,
|
| 318 |
+
"logits/chosen": -0.2059316188097,
|
| 319 |
+
"logits/rejected": -0.2756357192993164,
|
| 320 |
+
"logps/chosen": -12.392863273620605,
|
| 321 |
+
"logps/rejected": -26.90511131286621,
|
| 322 |
+
"loss": 0.4051,
|
| 323 |
+
"rewards/accuracies": 0.793749988079071,
|
| 324 |
+
"rewards/chosen": 1.0431259870529175,
|
| 325 |
+
"rewards/margins": 1.3048280477523804,
|
| 326 |
+
"rewards/rejected": -0.26170212030410767,
|
| 327 |
+
"step": 180
|
| 328 |
+
},
|
| 329 |
+
{
|
| 330 |
+
"epoch": 3.3853006681514475,
|
| 331 |
+
"grad_norm": 6.2247819900512695,
|
| 332 |
+
"learning_rate": 4.177453569964925e-06,
|
| 333 |
+
"logits/chosen": -0.17799881100654602,
|
| 334 |
+
"logits/rejected": -0.24945221841335297,
|
| 335 |
+
"logps/chosen": -12.190529823303223,
|
| 336 |
+
"logps/rejected": -27.486194610595703,
|
| 337 |
+
"loss": 0.3954,
|
| 338 |
+
"rewards/accuracies": 0.800000011920929,
|
| 339 |
+
"rewards/chosen": 1.0829803943634033,
|
| 340 |
+
"rewards/margins": 1.3739397525787354,
|
| 341 |
+
"rewards/rejected": -0.29095926880836487,
|
| 342 |
+
"step": 190
|
| 343 |
+
},
|
| 344 |
+
{
|
| 345 |
+
"epoch": 3.5634743875278394,
|
| 346 |
+
"grad_norm": 2.581493377685547,
|
| 347 |
+
"learning_rate": 4.058724504646834e-06,
|
| 348 |
+
"logits/chosen": -0.17182210087776184,
|
| 349 |
+
"logits/rejected": -0.23636431992053986,
|
| 350 |
+
"logps/chosen": -12.724395751953125,
|
| 351 |
+
"logps/rejected": -27.663299560546875,
|
| 352 |
+
"loss": 0.4032,
|
| 353 |
+
"rewards/accuracies": 0.793749988079071,
|
| 354 |
+
"rewards/chosen": 1.032086968421936,
|
| 355 |
+
"rewards/margins": 1.3668633699417114,
|
| 356 |
+
"rewards/rejected": -0.3347764015197754,
|
| 357 |
+
"step": 200
|
| 358 |
+
},
|
| 359 |
+
{
|
| 360 |
+
"epoch": 3.5634743875278394,
|
| 361 |
+
"eval_logits/chosen": -0.16310954093933105,
|
| 362 |
+
"eval_logits/rejected": -0.23061738908290863,
|
| 363 |
+
"eval_logps/chosen": -13.548603057861328,
|
| 364 |
+
"eval_logps/rejected": -28.07323455810547,
|
| 365 |
+
"eval_loss": 0.4708513021469116,
|
| 366 |
+
"eval_rewards/accuracies": 0.8100000023841858,
|
| 367 |
+
"eval_rewards/chosen": 0.9602789282798767,
|
| 368 |
+
"eval_rewards/margins": 1.244660496711731,
|
| 369 |
+
"eval_rewards/rejected": -0.28438156843185425,
|
| 370 |
+
"eval_runtime": 9.734,
|
| 371 |
+
"eval_samples_per_second": 10.273,
|
| 372 |
+
"eval_steps_per_second": 5.137,
|
| 373 |
+
"step": 200
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 3.7416481069042318,
|
| 377 |
+
"grad_norm": 2.387537717819214,
|
| 378 |
+
"learning_rate": 3.933941090877615e-06,
|
| 379 |
+
"logits/chosen": -0.15226894617080688,
|
| 380 |
+
"logits/rejected": -0.23664240539073944,
|
| 381 |
+
"logps/chosen": -12.170495986938477,
|
| 382 |
+
"logps/rejected": -28.98239517211914,
|
| 383 |
+
"loss": 0.3887,
|
| 384 |
+
"rewards/accuracies": 0.8125,
|
| 385 |
+
"rewards/chosen": 1.1489580869674683,
|
| 386 |
+
"rewards/margins": 1.5401805639266968,
|
| 387 |
+
"rewards/rejected": -0.39122244715690613,
|
| 388 |
+
"step": 210
|
| 389 |
+
},
|
| 390 |
+
{
|
| 391 |
+
"epoch": 3.9198218262806237,
|
| 392 |
+
"grad_norm": 7.2560529708862305,
|
| 393 |
+
"learning_rate": 3.8035880084487454e-06,
|
| 394 |
+
"logits/chosen": -0.19596333801746368,
|
| 395 |
+
"logits/rejected": -0.2662833034992218,
|
| 396 |
+
"logps/chosen": -13.05119514465332,
|
| 397 |
+
"logps/rejected": -29.070032119750977,
|
| 398 |
+
"loss": 0.4093,
|
| 399 |
+
"rewards/accuracies": 0.8125,
|
| 400 |
+
"rewards/chosen": 1.0079821348190308,
|
| 401 |
+
"rewards/margins": 1.3972289562225342,
|
| 402 |
+
"rewards/rejected": -0.38924697041511536,
|
| 403 |
+
"step": 220
|
| 404 |
+
},
|
| 405 |
+
{
|
| 406 |
+
"epoch": 4.097995545657016,
|
| 407 |
+
"grad_norm": 8.643023490905762,
|
| 408 |
+
"learning_rate": 3.6681715706826555e-06,
|
| 409 |
+
"logits/chosen": -0.1443440169095993,
|
| 410 |
+
"logits/rejected": -0.2094365656375885,
|
| 411 |
+
"logps/chosen": -12.210596084594727,
|
| 412 |
+
"logps/rejected": -28.835458755493164,
|
| 413 |
+
"loss": 0.3526,
|
| 414 |
+
"rewards/accuracies": 0.831250011920929,
|
| 415 |
+
"rewards/chosen": 1.1435935497283936,
|
| 416 |
+
"rewards/margins": 1.6025713682174683,
|
| 417 |
+
"rewards/rejected": -0.45897769927978516,
|
| 418 |
+
"step": 230
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 4.276169265033408,
|
| 422 |
+
"grad_norm": 8.177175521850586,
|
| 423 |
+
"learning_rate": 3.5282177578265295e-06,
|
| 424 |
+
"logits/chosen": -0.16080088913440704,
|
| 425 |
+
"logits/rejected": -0.22377625107765198,
|
| 426 |
+
"logps/chosen": -12.389388084411621,
|
| 427 |
+
"logps/rejected": -29.504175186157227,
|
| 428 |
+
"loss": 0.3327,
|
| 429 |
+
"rewards/accuracies": 0.862500011920929,
|
| 430 |
+
"rewards/chosen": 1.1701444387435913,
|
| 431 |
+
"rewards/margins": 1.6151951551437378,
|
| 432 |
+
"rewards/rejected": -0.44505080580711365,
|
| 433 |
+
"step": 240
|
| 434 |
+
},
|
| 435 |
+
{
|
| 436 |
+
"epoch": 4.4543429844097995,
|
| 437 |
+
"grad_norm": 3.6266558170318604,
|
| 438 |
+
"learning_rate": 3.384270174056454e-06,
|
| 439 |
+
"logits/chosen": -0.17858387529850006,
|
| 440 |
+
"logits/rejected": -0.23888497054576874,
|
| 441 |
+
"logps/chosen": -12.181685447692871,
|
| 442 |
+
"logps/rejected": -29.275564193725586,
|
| 443 |
+
"loss": 0.3836,
|
| 444 |
+
"rewards/accuracies": 0.8187500238418579,
|
| 445 |
+
"rewards/chosen": 1.0620548725128174,
|
| 446 |
+
"rewards/margins": 1.5715227127075195,
|
| 447 |
+
"rewards/rejected": -0.5094677805900574,
|
| 448 |
+
"step": 250
|
| 449 |
+
},
|
| 450 |
+
{
|
| 451 |
+
"epoch": 4.4543429844097995,
|
| 452 |
+
"eval_logits/chosen": -0.13963328301906586,
|
| 453 |
+
"eval_logits/rejected": -0.2079545557498932,
|
| 454 |
+
"eval_logps/chosen": -13.248819351196289,
|
| 455 |
+
"eval_logps/rejected": -29.226905822753906,
|
| 456 |
+
"eval_loss": 0.4674791693687439,
|
| 457 |
+
"eval_rewards/accuracies": 0.7899999618530273,
|
| 458 |
+
"eval_rewards/chosen": 0.9902573823928833,
|
| 459 |
+
"eval_rewards/margins": 1.390006184577942,
|
| 460 |
+
"eval_rewards/rejected": -0.3997488021850586,
|
| 461 |
+
"eval_runtime": 9.715,
|
| 462 |
+
"eval_samples_per_second": 10.293,
|
| 463 |
+
"eval_steps_per_second": 5.147,
|
| 464 |
+
"step": 250
|
| 465 |
+
},
|
| 466 |
+
{
|
| 467 |
+
"epoch": 4.632516703786192,
|
| 468 |
+
"grad_norm": 4.119279384613037,
|
| 469 |
+
"learning_rate": 3.236887936027261e-06,
|
| 470 |
+
"logits/chosen": -0.12825655937194824,
|
| 471 |
+
"logits/rejected": -0.19981206953525543,
|
| 472 |
+
"logps/chosen": -11.074287414550781,
|
| 473 |
+
"logps/rejected": -31.082468032836914,
|
| 474 |
+
"loss": 0.3368,
|
| 475 |
+
"rewards/accuracies": 0.862500011920929,
|
| 476 |
+
"rewards/chosen": 1.162680983543396,
|
| 477 |
+
"rewards/margins": 1.7504284381866455,
|
| 478 |
+
"rewards/rejected": -0.5877474546432495,
|
| 479 |
+
"step": 260
|
| 480 |
+
},
|
| 481 |
+
{
|
| 482 |
+
"epoch": 4.810690423162583,
|
| 483 |
+
"grad_norm": 6.509561061859131,
|
| 484 |
+
"learning_rate": 3.0866435011692884e-06,
|
| 485 |
+
"logits/chosen": -0.15560267865657806,
|
| 486 |
+
"logits/rejected": -0.2208879441022873,
|
| 487 |
+
"logps/chosen": -11.37994384765625,
|
| 488 |
+
"logps/rejected": -31.601425170898438,
|
| 489 |
+
"loss": 0.338,
|
| 490 |
+
"rewards/accuracies": 0.8812500238418579,
|
| 491 |
+
"rewards/chosen": 1.1841315031051636,
|
| 492 |
+
"rewards/margins": 1.8628618717193604,
|
| 493 |
+
"rewards/rejected": -0.6787301898002625,
|
| 494 |
+
"step": 270
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 4.988864142538976,
|
| 498 |
+
"grad_norm": 4.498559951782227,
|
| 499 |
+
"learning_rate": 2.9341204441673267e-06,
|
| 500 |
+
"logits/chosen": -0.1727777123451233,
|
| 501 |
+
"logits/rejected": -0.2310466766357422,
|
| 502 |
+
"logps/chosen": -13.475479125976562,
|
| 503 |
+
"logps/rejected": -27.789844512939453,
|
| 504 |
+
"loss": 0.5027,
|
| 505 |
+
"rewards/accuracies": 0.768750011920929,
|
| 506 |
+
"rewards/chosen": 0.9239163398742676,
|
| 507 |
+
"rewards/margins": 1.258786678314209,
|
| 508 |
+
"rewards/rejected": -0.33487027883529663,
|
| 509 |
+
"step": 280
|
| 510 |
+
},
|
| 511 |
+
{
|
| 512 |
+
"epoch": 5.167037861915367,
|
| 513 |
+
"grad_norm": 3.030965805053711,
|
| 514 |
+
"learning_rate": 2.7799111902582697e-06,
|
| 515 |
+
"logits/chosen": -0.15166734158992767,
|
| 516 |
+
"logits/rejected": -0.22125867009162903,
|
| 517 |
+
"logps/chosen": -12.045249938964844,
|
| 518 |
+
"logps/rejected": -31.249042510986328,
|
| 519 |
+
"loss": 0.32,
|
| 520 |
+
"rewards/accuracies": 0.856249988079071,
|
| 521 |
+
"rewards/chosen": 1.147273063659668,
|
| 522 |
+
"rewards/margins": 1.7833982706069946,
|
| 523 |
+
"rewards/rejected": -0.6361253261566162,
|
| 524 |
+
"step": 290
|
| 525 |
+
},
|
| 526 |
+
{
|
| 527 |
+
"epoch": 5.3452115812917596,
|
| 528 |
+
"grad_norm": 7.184595584869385,
|
| 529 |
+
"learning_rate": 2.624614714151743e-06,
|
| 530 |
+
"logits/chosen": -0.0994877815246582,
|
| 531 |
+
"logits/rejected": -0.17965565621852875,
|
| 532 |
+
"logps/chosen": -12.375929832458496,
|
| 533 |
+
"logps/rejected": -31.211719512939453,
|
| 534 |
+
"loss": 0.3588,
|
| 535 |
+
"rewards/accuracies": 0.8500000238418579,
|
| 536 |
+
"rewards/chosen": 1.1141376495361328,
|
| 537 |
+
"rewards/margins": 1.7681716680526733,
|
| 538 |
+
"rewards/rejected": -0.6540343165397644,
|
| 539 |
+
"step": 300
|
| 540 |
+
},
|
| 541 |
+
{
|
| 542 |
+
"epoch": 5.3452115812917596,
|
| 543 |
+
"eval_logits/chosen": -0.1254928857088089,
|
| 544 |
+
"eval_logits/rejected": -0.19310244917869568,
|
| 545 |
+
"eval_logps/chosen": -13.40658187866211,
|
| 546 |
+
"eval_logps/rejected": -29.754491806030273,
|
| 547 |
+
"eval_loss": 0.4751954674720764,
|
| 548 |
+
"eval_rewards/accuracies": 0.7699999809265137,
|
| 549 |
+
"eval_rewards/chosen": 0.9744812846183777,
|
| 550 |
+
"eval_rewards/margins": 1.4269884824752808,
|
| 551 |
+
"eval_rewards/rejected": -0.452507346868515,
|
| 552 |
+
"eval_runtime": 9.7087,
|
| 553 |
+
"eval_samples_per_second": 10.3,
|
| 554 |
+
"eval_steps_per_second": 5.15,
|
| 555 |
+
"step": 300
|
| 556 |
+
},
|
| 557 |
+
{
|
| 558 |
+
"epoch": 5.523385300668151,
|
| 559 |
+
"grad_norm": 3.5659422874450684,
|
| 560 |
+
"learning_rate": 2.4688342135114625e-06,
|
| 561 |
+
"logits/chosen": -0.13226279616355896,
|
| 562 |
+
"logits/rejected": -0.21120555698871613,
|
| 563 |
+
"logps/chosen": -11.256922721862793,
|
| 564 |
+
"logps/rejected": -30.438766479492188,
|
| 565 |
+
"loss": 0.3473,
|
| 566 |
+
"rewards/accuracies": 0.831250011920929,
|
| 567 |
+
"rewards/chosen": 1.2133493423461914,
|
| 568 |
+
"rewards/margins": 1.790173888206482,
|
| 569 |
+
"rewards/rejected": -0.5768246054649353,
|
| 570 |
+
"step": 310
|
| 571 |
+
},
|
| 572 |
+
{
|
| 573 |
+
"epoch": 5.701559020044543,
|
| 574 |
+
"grad_norm": 1.8460242748260498,
|
| 575 |
+
"learning_rate": 2.3131747660339396e-06,
|
| 576 |
+
"logits/chosen": -0.14635451138019562,
|
| 577 |
+
"logits/rejected": -0.19989195466041565,
|
| 578 |
+
"logps/chosen": -12.754526138305664,
|
| 579 |
+
"logps/rejected": -30.52984619140625,
|
| 580 |
+
"loss": 0.3378,
|
| 581 |
+
"rewards/accuracies": 0.856249988079071,
|
| 582 |
+
"rewards/chosen": 1.090704321861267,
|
| 583 |
+
"rewards/margins": 1.7227219343185425,
|
| 584 |
+
"rewards/rejected": -0.6320176124572754,
|
| 585 |
+
"step": 320
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 5.879732739420936,
|
| 589 |
+
"grad_norm": 5.234955310821533,
|
| 590 |
+
"learning_rate": 2.158240979224817e-06,
|
| 591 |
+
"logits/chosen": -0.1211402416229248,
|
| 592 |
+
"logits/rejected": -0.19876351952552795,
|
| 593 |
+
"logps/chosen": -10.352498054504395,
|
| 594 |
+
"logps/rejected": -31.713010787963867,
|
| 595 |
+
"loss": 0.3241,
|
| 596 |
+
"rewards/accuracies": 0.8812500238418579,
|
| 597 |
+
"rewards/chosen": 1.1969298124313354,
|
| 598 |
+
"rewards/margins": 1.8751522302627563,
|
| 599 |
+
"rewards/rejected": -0.6782223582267761,
|
| 600 |
+
"step": 330
|
| 601 |
+
},
|
| 602 |
+
{
|
| 603 |
+
"epoch": 6.057906458797327,
|
| 604 |
+
"grad_norm": 4.994526386260986,
|
| 605 |
+
"learning_rate": 2.004634642001507e-06,
|
| 606 |
+
"logits/chosen": -0.13834324479103088,
|
| 607 |
+
"logits/rejected": -0.2308368682861328,
|
| 608 |
+
"logps/chosen": -11.07118034362793,
|
| 609 |
+
"logps/rejected": -31.32500648498535,
|
| 610 |
+
"loss": 0.3868,
|
| 611 |
+
"rewards/accuracies": 0.8500000238418579,
|
| 612 |
+
"rewards/chosen": 1.1334035396575928,
|
| 613 |
+
"rewards/margins": 1.7996925115585327,
|
| 614 |
+
"rewards/rejected": -0.6662889719009399,
|
| 615 |
+
"step": 340
|
| 616 |
+
},
|
| 617 |
+
{
|
| 618 |
+
"epoch": 6.23608017817372,
|
| 619 |
+
"grad_norm": 3.519150733947754,
|
| 620 |
+
"learning_rate": 1.852952387243698e-06,
|
| 621 |
+
"logits/chosen": -0.1219867691397667,
|
| 622 |
+
"logits/rejected": -0.19822999835014343,
|
| 623 |
+
"logps/chosen": -10.62753677368164,
|
| 624 |
+
"logps/rejected": -32.93171310424805,
|
| 625 |
+
"loss": 0.2861,
|
| 626 |
+
"rewards/accuracies": 0.893750011920929,
|
| 627 |
+
"rewards/chosen": 1.1698468923568726,
|
| 628 |
+
"rewards/margins": 2.0219147205352783,
|
| 629 |
+
"rewards/rejected": -0.8520679473876953,
|
| 630 |
+
"step": 350
|
| 631 |
+
},
|
| 632 |
+
{
|
| 633 |
+
"epoch": 6.23608017817372,
|
| 634 |
+
"eval_logits/chosen": -0.11021164059638977,
|
| 635 |
+
"eval_logits/rejected": -0.17845386266708374,
|
| 636 |
+
"eval_logps/chosen": -13.759096145629883,
|
| 637 |
+
"eval_logps/rejected": -30.731969833374023,
|
| 638 |
+
"eval_loss": 0.4811996519565582,
|
| 639 |
+
"eval_rewards/accuracies": 0.7699999809265137,
|
| 640 |
+
"eval_rewards/chosen": 0.9392297267913818,
|
| 641 |
+
"eval_rewards/margins": 1.4894847869873047,
|
| 642 |
+
"eval_rewards/rejected": -0.5502550601959229,
|
| 643 |
+
"eval_runtime": 9.6948,
|
| 644 |
+
"eval_samples_per_second": 10.315,
|
| 645 |
+
"eval_steps_per_second": 5.157,
|
| 646 |
+
"step": 350
|
| 647 |
+
},
|
| 648 |
+
{
|
| 649 |
+
"epoch": 6.414253897550111,
|
| 650 |
+
"grad_norm": 2.1364710330963135,
|
| 651 |
+
"learning_rate": 1.7037833743707892e-06,
|
| 652 |
+
"logits/chosen": -0.13026945292949677,
|
| 653 |
+
"logits/rejected": -0.2250211238861084,
|
| 654 |
+
"logps/chosen": -11.657172203063965,
|
| 655 |
+
"logps/rejected": -31.794097900390625,
|
| 656 |
+
"loss": 0.3096,
|
| 657 |
+
"rewards/accuracies": 0.875,
|
| 658 |
+
"rewards/chosen": 1.1625641584396362,
|
| 659 |
+
"rewards/margins": 1.8979129791259766,
|
| 660 |
+
"rewards/rejected": -0.7353487610816956,
|
| 661 |
+
"step": 360
|
| 662 |
+
},
|
| 663 |
+
{
|
| 664 |
+
"epoch": 6.5924276169265035,
|
| 665 |
+
"grad_norm": 6.27597713470459,
|
| 666 |
+
"learning_rate": 1.5577070009474872e-06,
|
| 667 |
+
"logits/chosen": -0.08007471263408661,
|
| 668 |
+
"logits/rejected": -0.14471502602100372,
|
| 669 |
+
"logps/chosen": -12.703398704528809,
|
| 670 |
+
"logps/rejected": -31.92864418029785,
|
| 671 |
+
"loss": 0.3395,
|
| 672 |
+
"rewards/accuracies": 0.84375,
|
| 673 |
+
"rewards/chosen": 1.0662773847579956,
|
| 674 |
+
"rewards/margins": 1.84355628490448,
|
| 675 |
+
"rewards/rejected": -0.7772787809371948,
|
| 676 |
+
"step": 370
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 6.770601336302895,
|
| 680 |
+
"grad_norm": 1.5935697555541992,
|
| 681 |
+
"learning_rate": 1.415290652206105e-06,
|
| 682 |
+
"logits/chosen": -0.15049786865711212,
|
| 683 |
+
"logits/rejected": -0.21002641320228577,
|
| 684 |
+
"logps/chosen": -12.739409446716309,
|
| 685 |
+
"logps/rejected": -32.949214935302734,
|
| 686 |
+
"loss": 0.3588,
|
| 687 |
+
"rewards/accuracies": 0.84375,
|
| 688 |
+
"rewards/chosen": 1.0807316303253174,
|
| 689 |
+
"rewards/margins": 1.8172613382339478,
|
| 690 |
+
"rewards/rejected": -0.7365297675132751,
|
| 691 |
+
"step": 380
|
| 692 |
+
},
|
| 693 |
+
{
|
| 694 |
+
"epoch": 6.948775055679287,
|
| 695 |
+
"grad_norm": 12.438933372497559,
|
| 696 |
+
"learning_rate": 1.2770874972267777e-06,
|
| 697 |
+
"logits/chosen": -0.10974359512329102,
|
| 698 |
+
"logits/rejected": -0.1568942815065384,
|
| 699 |
+
"logps/chosen": -12.335737228393555,
|
| 700 |
+
"logps/rejected": -33.184993743896484,
|
| 701 |
+
"loss": 0.3357,
|
| 702 |
+
"rewards/accuracies": 0.8687500357627869,
|
| 703 |
+
"rewards/chosen": 1.1363105773925781,
|
| 704 |
+
"rewards/margins": 1.9653358459472656,
|
| 705 |
+
"rewards/rejected": -0.8290252685546875,
|
| 706 |
+
"step": 390
|
| 707 |
+
},
|
| 708 |
+
{
|
| 709 |
+
"epoch": 7.12694877505568,
|
| 710 |
+
"grad_norm": 2.4859490394592285,
|
| 711 |
+
"learning_rate": 1.1436343403356019e-06,
|
| 712 |
+
"logits/chosen": -0.11985152959823608,
|
| 713 |
+
"logits/rejected": -0.18024906516075134,
|
| 714 |
+
"logps/chosen": -13.101457595825195,
|
| 715 |
+
"logps/rejected": -32.767181396484375,
|
| 716 |
+
"loss": 0.3662,
|
| 717 |
+
"rewards/accuracies": 0.84375,
|
| 718 |
+
"rewards/chosen": 0.9988664984703064,
|
| 719 |
+
"rewards/margins": 1.8207927942276,
|
| 720 |
+
"rewards/rejected": -0.8219264149665833,
|
| 721 |
+
"step": 400
|
| 722 |
+
},
|
| 723 |
+
{
|
| 724 |
+
"epoch": 7.12694877505568,
|
| 725 |
+
"eval_logits/chosen": -0.099003367125988,
|
| 726 |
+
"eval_logits/rejected": -0.16790008544921875,
|
| 727 |
+
"eval_logps/chosen": -13.986225128173828,
|
| 728 |
+
"eval_logps/rejected": -31.5858154296875,
|
| 729 |
+
"eval_loss": 0.4867512881755829,
|
| 730 |
+
"eval_rewards/accuracies": 0.7699999809265137,
|
| 731 |
+
"eval_rewards/chosen": 0.9165167212486267,
|
| 732 |
+
"eval_rewards/margins": 1.5521563291549683,
|
| 733 |
+
"eval_rewards/rejected": -0.6356395483016968,
|
| 734 |
+
"eval_runtime": 9.7105,
|
| 735 |
+
"eval_samples_per_second": 10.298,
|
| 736 |
+
"eval_steps_per_second": 5.149,
|
| 737 |
+
"step": 400
|
| 738 |
+
},
|
| 739 |
+
{
|
| 740 |
+
"epoch": 7.305122494432071,
|
| 741 |
+
"grad_norm": 4.4709153175354,
|
| 742 |
+
"learning_rate": 1.0154495360662464e-06,
|
| 743 |
+
"logits/chosen": -0.12129449844360352,
|
| 744 |
+
"logits/rejected": -0.18251024186611176,
|
| 745 |
+
"logps/chosen": -13.18463134765625,
|
| 746 |
+
"logps/rejected": -32.92716598510742,
|
| 747 |
+
"loss": 0.366,
|
| 748 |
+
"rewards/accuracies": 0.862500011920929,
|
| 749 |
+
"rewards/chosen": 1.057719349861145,
|
| 750 |
+
"rewards/margins": 1.885604739189148,
|
| 751 |
+
"rewards/rejected": -0.8278852701187134,
|
| 752 |
+
"step": 410
|
| 753 |
+
},
|
| 754 |
+
{
|
| 755 |
+
"epoch": 7.4832962138084635,
|
| 756 |
+
"grad_norm": 3.132544994354248,
|
| 757 |
+
"learning_rate": 8.930309757836517e-07,
|
| 758 |
+
"logits/chosen": -0.10683136433362961,
|
| 759 |
+
"logits/rejected": -0.1750011295080185,
|
| 760 |
+
"logps/chosen": -11.23306941986084,
|
| 761 |
+
"logps/rejected": -32.81039047241211,
|
| 762 |
+
"loss": 0.3044,
|
| 763 |
+
"rewards/accuracies": 0.8812500238418579,
|
| 764 |
+
"rewards/chosen": 1.1720637083053589,
|
| 765 |
+
"rewards/margins": 2.0555508136749268,
|
| 766 |
+
"rewards/rejected": -0.8834872245788574,
|
| 767 |
+
"step": 420
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 7.661469933184855,
|
| 771 |
+
"grad_norm": 6.80822229385376,
|
| 772 |
+
"learning_rate": 7.768541537901325e-07,
|
| 773 |
+
"logits/chosen": -0.11390836536884308,
|
| 774 |
+
"logits/rejected": -0.2062007039785385,
|
| 775 |
+
"logps/chosen": -11.768353462219238,
|
| 776 |
+
"logps/rejected": -33.399192810058594,
|
| 777 |
+
"loss": 0.3337,
|
| 778 |
+
"rewards/accuracies": 0.8687500357627869,
|
| 779 |
+
"rewards/chosen": 1.1444917917251587,
|
| 780 |
+
"rewards/margins": 1.9670883417129517,
|
| 781 |
+
"rewards/rejected": -0.8225963711738586,
|
| 782 |
+
"step": 430
|
| 783 |
+
},
|
| 784 |
+
{
|
| 785 |
+
"epoch": 7.839643652561247,
|
| 786 |
+
"grad_norm": 4.160062789916992,
|
| 787 |
+
"learning_rate": 6.673703204254348e-07,
|
| 788 |
+
"logits/chosen": -0.1045166477560997,
|
| 789 |
+
"logits/rejected": -0.18522079288959503,
|
| 790 |
+
"logps/chosen": -10.534876823425293,
|
| 791 |
+
"logps/rejected": -34.59806442260742,
|
| 792 |
+
"loss": 0.2644,
|
| 793 |
+
"rewards/accuracies": 0.893750011920929,
|
| 794 |
+
"rewards/chosen": 1.2115627527236938,
|
| 795 |
+
"rewards/margins": 2.1745359897613525,
|
| 796 |
+
"rewards/rejected": -0.9629732370376587,
|
| 797 |
+
"step": 440
|
| 798 |
+
},
|
| 799 |
+
{
|
| 800 |
+
"epoch": 8.017817371937639,
|
| 801 |
+
"grad_norm": 2.6154518127441406,
|
| 802 |
+
"learning_rate": 5.650047293344316e-07,
|
| 803 |
+
"logits/chosen": -0.07358330488204956,
|
| 804 |
+
"logits/rejected": -0.14352913200855255,
|
| 805 |
+
"logps/chosen": -10.867822647094727,
|
| 806 |
+
"logps/rejected": -33.30778884887695,
|
| 807 |
+
"loss": 0.2822,
|
| 808 |
+
"rewards/accuracies": 0.8687500357627869,
|
| 809 |
+
"rewards/chosen": 1.2662678956985474,
|
| 810 |
+
"rewards/margins": 2.097227096557617,
|
| 811 |
+
"rewards/rejected": -0.830959141254425,
|
| 812 |
+
"step": 450
|
| 813 |
+
},
|
| 814 |
+
{
|
| 815 |
+
"epoch": 8.017817371937639,
|
| 816 |
+
"eval_logits/chosen": -0.09359554946422577,
|
| 817 |
+
"eval_logits/rejected": -0.16219539940357208,
|
| 818 |
+
"eval_logps/chosen": -14.051919937133789,
|
| 819 |
+
"eval_logps/rejected": -31.741615295410156,
|
| 820 |
+
"eval_loss": 0.49268820881843567,
|
| 821 |
+
"eval_rewards/accuracies": 0.7599999904632568,
|
| 822 |
+
"eval_rewards/chosen": 0.9099473357200623,
|
| 823 |
+
"eval_rewards/margins": 1.5611672401428223,
|
| 824 |
+
"eval_rewards/rejected": -0.6512197852134705,
|
| 825 |
+
"eval_runtime": 9.7252,
|
| 826 |
+
"eval_samples_per_second": 10.283,
|
| 827 |
+
"eval_steps_per_second": 5.141,
|
| 828 |
+
"step": 450
|
| 829 |
+
},
|
| 830 |
+
{
|
| 831 |
+
"epoch": 8.195991091314031,
|
| 832 |
+
"grad_norm": 3.175278425216675,
|
| 833 |
+
"learning_rate": 4.7015498571035877e-07,
|
| 834 |
+
"logits/chosen": -0.08536185324192047,
|
| 835 |
+
"logits/rejected": -0.16373853385448456,
|
| 836 |
+
"logps/chosen": -10.688807487487793,
|
| 837 |
+
"logps/rejected": -35.358089447021484,
|
| 838 |
+
"loss": 0.2417,
|
| 839 |
+
"rewards/accuracies": 0.925000011920929,
|
| 840 |
+
"rewards/chosen": 1.2969163656234741,
|
| 841 |
+
"rewards/margins": 2.3196523189544678,
|
| 842 |
+
"rewards/rejected": -1.022735834121704,
|
| 843 |
+
"step": 460
|
| 844 |
+
},
|
| 845 |
+
{
|
| 846 |
+
"epoch": 8.374164810690424,
|
| 847 |
+
"grad_norm": 10.432084083557129,
|
| 848 |
+
"learning_rate": 3.831895019292897e-07,
|
| 849 |
+
"logits/chosen": -0.11144526302814484,
|
| 850 |
+
"logits/rejected": -0.18342572450637817,
|
| 851 |
+
"logps/chosen": -12.896112442016602,
|
| 852 |
+
"logps/rejected": -34.08356857299805,
|
| 853 |
+
"loss": 0.3277,
|
| 854 |
+
"rewards/accuracies": 0.90625,
|
| 855 |
+
"rewards/chosen": 1.0476747751235962,
|
| 856 |
+
"rewards/margins": 1.9816349744796753,
|
| 857 |
+
"rewards/rejected": -0.9339599609375,
|
| 858 |
+
"step": 470
|
| 859 |
+
},
|
| 860 |
+
{
|
| 861 |
+
"epoch": 8.552338530066816,
|
| 862 |
+
"grad_norm": 4.943639278411865,
|
| 863 |
+
"learning_rate": 3.044460665744284e-07,
|
| 864 |
+
"logits/chosen": -0.06191817671060562,
|
| 865 |
+
"logits/rejected": -0.13474522531032562,
|
| 866 |
+
"logps/chosen": -11.490986824035645,
|
| 867 |
+
"logps/rejected": -33.3411750793457,
|
| 868 |
+
"loss": 0.3482,
|
| 869 |
+
"rewards/accuracies": 0.862500011920929,
|
| 870 |
+
"rewards/chosen": 1.0883612632751465,
|
| 871 |
+
"rewards/margins": 1.9929107427597046,
|
| 872 |
+
"rewards/rejected": -0.9045494198799133,
|
| 873 |
+
"step": 480
|
| 874 |
+
},
|
| 875 |
+
{
|
| 876 |
+
"epoch": 8.730512249443207,
|
| 877 |
+
"grad_norm": 4.994484901428223,
|
| 878 |
+
"learning_rate": 2.3423053240837518e-07,
|
| 879 |
+
"logits/chosen": -0.11047738045454025,
|
| 880 |
+
"logits/rejected": -0.1758051961660385,
|
| 881 |
+
"logps/chosen": -12.3035306930542,
|
| 882 |
+
"logps/rejected": -33.04331588745117,
|
| 883 |
+
"loss": 0.3433,
|
| 884 |
+
"rewards/accuracies": 0.831250011920929,
|
| 885 |
+
"rewards/chosen": 1.051175594329834,
|
| 886 |
+
"rewards/margins": 1.9459151029586792,
|
| 887 |
+
"rewards/rejected": -0.8947394490242004,
|
| 888 |
+
"step": 490
|
| 889 |
+
},
|
| 890 |
+
{
|
| 891 |
+
"epoch": 8.908685968819599,
|
| 892 |
+
"grad_norm": 2.0751497745513916,
|
| 893 |
+
"learning_rate": 1.7281562838948968e-07,
|
| 894 |
+
"logits/chosen": -0.12217041105031967,
|
| 895 |
+
"logits/rejected": -0.19585369527339935,
|
| 896 |
+
"logps/chosen": -10.848115921020508,
|
| 897 |
+
"logps/rejected": -34.68525314331055,
|
| 898 |
+
"loss": 0.2416,
|
| 899 |
+
"rewards/accuracies": 0.887499988079071,
|
| 900 |
+
"rewards/chosen": 1.2858046293258667,
|
| 901 |
+
"rewards/margins": 2.253690481185913,
|
| 902 |
+
"rewards/rejected": -0.9678859114646912,
|
| 903 |
+
"step": 500
|
| 904 |
+
},
|
| 905 |
+
{
|
| 906 |
+
"epoch": 8.908685968819599,
|
| 907 |
+
"eval_logits/chosen": -0.08983445167541504,
|
| 908 |
+
"eval_logits/rejected": -0.15845781564712524,
|
| 909 |
+
"eval_logps/chosen": -14.23983097076416,
|
| 910 |
+
"eval_logps/rejected": -32.18784713745117,
|
| 911 |
+
"eval_loss": 0.49790748953819275,
|
| 912 |
+
"eval_rewards/accuracies": 0.7599999904632568,
|
| 913 |
+
"eval_rewards/chosen": 0.8911561369895935,
|
| 914 |
+
"eval_rewards/margins": 1.586998701095581,
|
| 915 |
+
"eval_rewards/rejected": -0.6958425641059875,
|
| 916 |
+
"eval_runtime": 9.7042,
|
| 917 |
+
"eval_samples_per_second": 10.305,
|
| 918 |
+
"eval_steps_per_second": 5.152,
|
| 919 |
+
"step": 500
|
| 920 |
+
},
|
| 921 |
+
{
|
| 922 |
+
"epoch": 9.086859688195991,
|
| 923 |
+
"grad_norm": 4.091846466064453,
|
| 924 |
+
"learning_rate": 1.2043990034669413e-07,
|
| 925 |
+
"logits/chosen": -0.10412635654211044,
|
| 926 |
+
"logits/rejected": -0.20009151101112366,
|
| 927 |
+
"logps/chosen": -11.962130546569824,
|
| 928 |
+
"logps/rejected": -34.19890594482422,
|
| 929 |
+
"loss": 0.3145,
|
| 930 |
+
"rewards/accuracies": 0.887499988079071,
|
| 931 |
+
"rewards/chosen": 1.1489673852920532,
|
| 932 |
+
"rewards/margins": 2.0444374084472656,
|
| 933 |
+
"rewards/rejected": -0.8954699635505676,
|
| 934 |
+
"step": 510
|
| 935 |
+
},
|
| 936 |
+
{
|
| 937 |
+
"epoch": 9.265033407572384,
|
| 938 |
+
"grad_norm": 2.8685364723205566,
|
| 939 |
+
"learning_rate": 7.730678442730539e-08,
|
| 940 |
+
"logits/chosen": -0.13043461740016937,
|
| 941 |
+
"logits/rejected": -0.1936294585466385,
|
| 942 |
+
"logps/chosen": -12.278414726257324,
|
| 943 |
+
"logps/rejected": -35.482120513916016,
|
| 944 |
+
"loss": 0.2503,
|
| 945 |
+
"rewards/accuracies": 0.90625,
|
| 946 |
+
"rewards/chosen": 1.113724946975708,
|
| 947 |
+
"rewards/margins": 2.1855549812316895,
|
| 948 |
+
"rewards/rejected": -1.0718300342559814,
|
| 949 |
+
"step": 520
|
| 950 |
+
},
|
| 951 |
+
{
|
| 952 |
+
"epoch": 9.443207126948774,
|
| 953 |
+
"grad_norm": 8.583150863647461,
|
| 954 |
+
"learning_rate": 4.358381691677932e-08,
|
| 955 |
+
"logits/chosen": -0.1350499838590622,
|
| 956 |
+
"logits/rejected": -0.1929396241903305,
|
| 957 |
+
"logps/chosen": -12.875258445739746,
|
| 958 |
+
"logps/rejected": -33.197200775146484,
|
| 959 |
+
"loss": 0.3269,
|
| 960 |
+
"rewards/accuracies": 0.862500011920929,
|
| 961 |
+
"rewards/chosen": 1.0666710138320923,
|
| 962 |
+
"rewards/margins": 1.9351733922958374,
|
| 963 |
+
"rewards/rejected": -0.8685024380683899,
|
| 964 |
+
"step": 530
|
| 965 |
+
},
|
| 966 |
+
{
|
| 967 |
+
"epoch": 9.621380846325167,
|
| 968 |
+
"grad_norm": 11.428977012634277,
|
| 969 |
+
"learning_rate": 1.9401983499569843e-08,
|
| 970 |
+
"logits/chosen": -0.1044853925704956,
|
| 971 |
+
"logits/rejected": -0.17102347314357758,
|
| 972 |
+
"logps/chosen": -11.800514221191406,
|
| 973 |
+
"logps/rejected": -33.656124114990234,
|
| 974 |
+
"loss": 0.3398,
|
| 975 |
+
"rewards/accuracies": 0.856249988079071,
|
| 976 |
+
"rewards/chosen": 1.1207963228225708,
|
| 977 |
+
"rewards/margins": 1.9993568658828735,
|
| 978 |
+
"rewards/rejected": -0.878560483455658,
|
| 979 |
+
"step": 540
|
| 980 |
+
},
|
| 981 |
+
{
|
| 982 |
+
"epoch": 9.799554565701559,
|
| 983 |
+
"grad_norm": 2.2846410274505615,
|
| 984 |
+
"learning_rate": 4.855210488670381e-09,
|
| 985 |
+
"logits/chosen": -0.08743849396705627,
|
| 986 |
+
"logits/rejected": -0.16115520894527435,
|
| 987 |
+
"logps/chosen": -11.04604434967041,
|
| 988 |
+
"logps/rejected": -34.041011810302734,
|
| 989 |
+
"loss": 0.3096,
|
| 990 |
+
"rewards/accuracies": 0.887499988079071,
|
| 991 |
+
"rewards/chosen": 1.1628315448760986,
|
| 992 |
+
"rewards/margins": 2.1141316890716553,
|
| 993 |
+
"rewards/rejected": -0.9513001441955566,
|
| 994 |
+
"step": 550
|
| 995 |
+
},
|
| 996 |
+
{
|
| 997 |
+
"epoch": 9.799554565701559,
|
| 998 |
+
"eval_logits/chosen": -0.08727238327264786,
|
| 999 |
+
"eval_logits/rejected": -0.15482930839061737,
|
| 1000 |
+
"eval_logps/chosen": -14.208100318908691,
|
| 1001 |
+
"eval_logps/rejected": -32.24625015258789,
|
| 1002 |
+
"eval_loss": 0.4933530390262604,
|
| 1003 |
+
"eval_rewards/accuracies": 0.75,
|
| 1004 |
+
"eval_rewards/chosen": 0.8943293690681458,
|
| 1005 |
+
"eval_rewards/margins": 1.5960127115249634,
|
| 1006 |
+
"eval_rewards/rejected": -0.7016833424568176,
|
| 1007 |
+
"eval_runtime": 9.7018,
|
| 1008 |
+
"eval_samples_per_second": 10.307,
|
| 1009 |
+
"eval_steps_per_second": 5.154,
|
| 1010 |
+
"step": 550
|
| 1011 |
+
},
|
| 1012 |
+
{
|
| 1013 |
+
"epoch": 9.977728285077951,
|
| 1014 |
+
"grad_norm": 2.3474748134613037,
|
| 1015 |
+
"learning_rate": 0.0,
|
| 1016 |
+
"logits/chosen": -0.030874544754624367,
|
| 1017 |
+
"logits/rejected": -0.118435338139534,
|
| 1018 |
+
"logps/chosen": -10.919882774353027,
|
| 1019 |
+
"logps/rejected": -32.95122146606445,
|
| 1020 |
+
"loss": 0.3123,
|
| 1021 |
+
"rewards/accuracies": 0.8500000238418579,
|
| 1022 |
+
"rewards/chosen": 1.201145052909851,
|
| 1023 |
+
"rewards/margins": 2.0579233169555664,
|
| 1024 |
+
"rewards/rejected": -0.8567783236503601,
|
| 1025 |
+
"step": 560
|
| 1026 |
+
},
|
| 1027 |
+
{
|
| 1028 |
+
"epoch": 9.977728285077951,
|
| 1029 |
+
"step": 560,
|
| 1030 |
+
"total_flos": 5.23282185018409e+16,
|
| 1031 |
+
"train_loss": 0.4115582968507494,
|
| 1032 |
+
"train_runtime": 2017.6604,
|
| 1033 |
+
"train_samples_per_second": 4.451,
|
| 1034 |
+
"train_steps_per_second": 0.278
|
| 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": 5.23282185018409e+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
|