End of training
Browse files- README.md +3 -2
- all_results.json +20 -0
- eval_results.json +15 -0
- train_results.json +8 -0
- trainer_state.json +861 -0
- training_eval_loss.png +0 -0
- training_loss.png +0 -0
- training_rewards_chosen.png +0 -0
README.md
CHANGED
|
@@ -3,9 +3,10 @@ library_name: peft
|
|
| 3 |
license: llama3.1
|
| 4 |
base_model: meta-llama/Meta-Llama-3.1-8B-Instruct
|
| 5 |
tags:
|
|
|
|
|
|
|
| 6 |
- trl
|
| 7 |
- kto
|
| 8 |
-
- llama-factory
|
| 9 |
- generated_from_trainer
|
| 10 |
model-index:
|
| 11 |
- name: Llama-3.1-8B-Instruct-KTO-400
|
|
@@ -17,7 +18,7 @@ should probably proofread and complete it, then remove this comment. -->
|
|
| 17 |
|
| 18 |
# Llama-3.1-8B-Instruct-KTO-400
|
| 19 |
|
| 20 |
-
This model is a fine-tuned version of [meta-llama/Meta-Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct) on
|
| 21 |
It achieves the following results on the evaluation set:
|
| 22 |
- Loss: 0.2541
|
| 23 |
- Rewards/chosen: 0.0309
|
|
|
|
| 3 |
license: llama3.1
|
| 4 |
base_model: meta-llama/Meta-Llama-3.1-8B-Instruct
|
| 5 |
tags:
|
| 6 |
+
- llama-factory
|
| 7 |
+
- lora
|
| 8 |
- trl
|
| 9 |
- kto
|
|
|
|
| 10 |
- generated_from_trainer
|
| 11 |
model-index:
|
| 12 |
- name: Llama-3.1-8B-Instruct-KTO-400
|
|
|
|
| 18 |
|
| 19 |
# Llama-3.1-8B-Instruct-KTO-400
|
| 20 |
|
| 21 |
+
This model is a fine-tuned version of [meta-llama/Meta-Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct) on the bct_non_cot_kto_400 dataset.
|
| 22 |
It achieves the following results on the evaluation set:
|
| 23 |
- Loss: 0.2541
|
| 24 |
- Rewards/chosen: 0.0309
|
all_results.json
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 10.0,
|
| 3 |
+
"eval_logits/chosen": -5221032.228571429,
|
| 4 |
+
"eval_logits/rejected": -5284203.377777778,
|
| 5 |
+
"eval_logps/chosen": -16.849844796316965,
|
| 6 |
+
"eval_logps/rejected": -62.744411892361114,
|
| 7 |
+
"eval_loss": 0.2541462779045105,
|
| 8 |
+
"eval_rewards/chosen": 0.030902576446533204,
|
| 9 |
+
"eval_rewards/margins": 4.341355186038547,
|
| 10 |
+
"eval_rewards/rejected": -4.310452609592014,
|
| 11 |
+
"eval_runtime": 17.1784,
|
| 12 |
+
"eval_samples_per_second": 4.657,
|
| 13 |
+
"eval_steps_per_second": 2.329,
|
| 14 |
+
"kl": 0.0,
|
| 15 |
+
"total_flos": 3.939427351343923e+16,
|
| 16 |
+
"train_loss": 0.30883768532011246,
|
| 17 |
+
"train_runtime": 2497.4829,
|
| 18 |
+
"train_samples_per_second": 2.883,
|
| 19 |
+
"train_steps_per_second": 0.18
|
| 20 |
+
}
|
eval_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 10.0,
|
| 3 |
+
"eval_logits/chosen": -5221032.228571429,
|
| 4 |
+
"eval_logits/rejected": -5284203.377777778,
|
| 5 |
+
"eval_logps/chosen": -16.849844796316965,
|
| 6 |
+
"eval_logps/rejected": -62.744411892361114,
|
| 7 |
+
"eval_loss": 0.2541462779045105,
|
| 8 |
+
"eval_rewards/chosen": 0.030902576446533204,
|
| 9 |
+
"eval_rewards/margins": 4.341355186038547,
|
| 10 |
+
"eval_rewards/rejected": -4.310452609592014,
|
| 11 |
+
"eval_runtime": 17.1784,
|
| 12 |
+
"eval_samples_per_second": 4.657,
|
| 13 |
+
"eval_steps_per_second": 2.329,
|
| 14 |
+
"kl": 0.0
|
| 15 |
+
}
|
train_results.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 10.0,
|
| 3 |
+
"total_flos": 3.939427351343923e+16,
|
| 4 |
+
"train_loss": 0.30883768532011246,
|
| 5 |
+
"train_runtime": 2497.4829,
|
| 6 |
+
"train_samples_per_second": 2.883,
|
| 7 |
+
"train_steps_per_second": 0.18
|
| 8 |
+
}
|
trainer_state.json
ADDED
|
@@ -0,0 +1,861 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": 0.2541462779045105,
|
| 3 |
+
"best_model_checkpoint": "saves/sycophancy/Llama-8B-3.1-Instruct/kto-400/checkpoint-450",
|
| 4 |
+
"epoch": 10.0,
|
| 5 |
+
"eval_steps": 50,
|
| 6 |
+
"global_step": 450,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.2222222222222222,
|
| 13 |
+
"grad_norm": 0.8755188584327698,
|
| 14 |
+
"kl": 3.4999122619628906,
|
| 15 |
+
"learning_rate": 1.111111111111111e-06,
|
| 16 |
+
"logits/chosen": -6424420.894117647,
|
| 17 |
+
"logits/rejected": -5984302.08,
|
| 18 |
+
"logps/chosen": -17.799721392463237,
|
| 19 |
+
"logps/rejected": -18.852799479166666,
|
| 20 |
+
"loss": 0.5001,
|
| 21 |
+
"rewards/chosen": -0.0005647187723832972,
|
| 22 |
+
"rewards/margins": 0.00044351912596646476,
|
| 23 |
+
"rewards/rejected": -0.001008237898349762,
|
| 24 |
+
"step": 10
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.4444444444444444,
|
| 28 |
+
"grad_norm": 0.924457311630249,
|
| 29 |
+
"kl": 5.80185604095459,
|
| 30 |
+
"learning_rate": 2.222222222222222e-06,
|
| 31 |
+
"logits/chosen": -6300276.363636363,
|
| 32 |
+
"logits/rejected": -8026647.111111111,
|
| 33 |
+
"logps/chosen": -17.250682484019887,
|
| 34 |
+
"logps/rejected": -21.154300265842014,
|
| 35 |
+
"loss": 0.501,
|
| 36 |
+
"rewards/chosen": -0.004935111511837353,
|
| 37 |
+
"rewards/margins": -0.007736933182435807,
|
| 38 |
+
"rewards/rejected": 0.002801821670598454,
|
| 39 |
+
"step": 20
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.6666666666666666,
|
| 43 |
+
"grad_norm": 0.6622861623764038,
|
| 44 |
+
"kl": 5.680029392242432,
|
| 45 |
+
"learning_rate": 3.3333333333333333e-06,
|
| 46 |
+
"logits/chosen": -6518788.860759494,
|
| 47 |
+
"logits/rejected": -7602240.790123457,
|
| 48 |
+
"logps/chosen": -17.539696029469937,
|
| 49 |
+
"logps/rejected": -20.295280550733025,
|
| 50 |
+
"loss": 0.5001,
|
| 51 |
+
"rewards/chosen": 0.004383341798299476,
|
| 52 |
+
"rewards/margins": 0.0011355700642266377,
|
| 53 |
+
"rewards/rejected": 0.003247771734072838,
|
| 54 |
+
"step": 30
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"epoch": 0.8888888888888888,
|
| 58 |
+
"grad_norm": 0.7431591153144836,
|
| 59 |
+
"kl": 6.487192630767822,
|
| 60 |
+
"learning_rate": 4.444444444444444e-06,
|
| 61 |
+
"logits/chosen": -5974229.333333333,
|
| 62 |
+
"logits/rejected": -7606852.363636363,
|
| 63 |
+
"logps/chosen": -16.686967637803818,
|
| 64 |
+
"logps/rejected": -18.974927035245027,
|
| 65 |
+
"loss": 0.4992,
|
| 66 |
+
"rewards/chosen": 0.0100140952401691,
|
| 67 |
+
"rewards/margins": 0.00425971449926646,
|
| 68 |
+
"rewards/rejected": 0.00575438074090264,
|
| 69 |
+
"step": 40
|
| 70 |
+
},
|
| 71 |
+
{
|
| 72 |
+
"epoch": 1.1111111111111112,
|
| 73 |
+
"grad_norm": 0.7480154633522034,
|
| 74 |
+
"kl": 5.691668510437012,
|
| 75 |
+
"learning_rate": 4.998119881260576e-06,
|
| 76 |
+
"logits/chosen": -5965018.285714285,
|
| 77 |
+
"logits/rejected": -7548874.947368421,
|
| 78 |
+
"logps/chosen": -17.54410371326265,
|
| 79 |
+
"logps/rejected": -21.369326942845394,
|
| 80 |
+
"loss": 0.4989,
|
| 81 |
+
"rewards/chosen": 0.018393808887118383,
|
| 82 |
+
"rewards/margins": 0.005255397810673055,
|
| 83 |
+
"rewards/rejected": 0.013138411076445328,
|
| 84 |
+
"step": 50
|
| 85 |
+
},
|
| 86 |
+
{
|
| 87 |
+
"epoch": 1.1111111111111112,
|
| 88 |
+
"eval_logits/chosen": -6615963.428571428,
|
| 89 |
+
"eval_logits/rejected": -7226443.377777778,
|
| 90 |
+
"eval_logps/chosen": -16.866085379464284,
|
| 91 |
+
"eval_logps/rejected": -19.409551323784722,
|
| 92 |
+
"eval_loss": 0.4995289444923401,
|
| 93 |
+
"eval_rewards/chosen": 0.02927844183785575,
|
| 94 |
+
"eval_rewards/margins": 0.006244459984794497,
|
| 95 |
+
"eval_rewards/rejected": 0.023033981853061252,
|
| 96 |
+
"eval_runtime": 17.4874,
|
| 97 |
+
"eval_samples_per_second": 4.575,
|
| 98 |
+
"eval_steps_per_second": 2.287,
|
| 99 |
+
"kl": 3.9760074615478516,
|
| 100 |
+
"step": 50
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 1.3333333333333333,
|
| 104 |
+
"grad_norm": 0.7908056378364563,
|
| 105 |
+
"kl": 6.587210178375244,
|
| 106 |
+
"learning_rate": 4.983095894354858e-06,
|
| 107 |
+
"logits/chosen": -6344583.441860465,
|
| 108 |
+
"logits/rejected": -7122381.405405405,
|
| 109 |
+
"logps/chosen": -16.889287904251454,
|
| 110 |
+
"logps/rejected": -20.947681323902028,
|
| 111 |
+
"loss": 0.4962,
|
| 112 |
+
"rewards/chosen": 0.045913998470749964,
|
| 113 |
+
"rewards/margins": 0.025308867695195206,
|
| 114 |
+
"rewards/rejected": 0.02060513077555476,
|
| 115 |
+
"step": 60
|
| 116 |
+
},
|
| 117 |
+
{
|
| 118 |
+
"epoch": 1.5555555555555556,
|
| 119 |
+
"grad_norm": 0.8089656233787537,
|
| 120 |
+
"kl": 12.094493865966797,
|
| 121 |
+
"learning_rate": 4.953138276568462e-06,
|
| 122 |
+
"logits/chosen": -5830952.205128205,
|
| 123 |
+
"logits/rejected": -7457981.658536585,
|
| 124 |
+
"logps/chosen": -17.15894493689904,
|
| 125 |
+
"logps/rejected": -19.85860089557927,
|
| 126 |
+
"loss": 0.4933,
|
| 127 |
+
"rewards/chosen": 0.09406566619873047,
|
| 128 |
+
"rewards/margins": 0.053329967870944885,
|
| 129 |
+
"rewards/rejected": 0.040735698327785584,
|
| 130 |
+
"step": 70
|
| 131 |
+
},
|
| 132 |
+
{
|
| 133 |
+
"epoch": 1.7777777777777777,
|
| 134 |
+
"grad_norm": 1.1325657367706299,
|
| 135 |
+
"kl": 13.553695678710938,
|
| 136 |
+
"learning_rate": 4.908427196539701e-06,
|
| 137 |
+
"logits/chosen": -5969041.090909091,
|
| 138 |
+
"logits/rejected": -8059607.111111111,
|
| 139 |
+
"logps/chosen": -15.58839277787642,
|
| 140 |
+
"logps/rejected": -18.098347981770832,
|
| 141 |
+
"loss": 0.484,
|
| 142 |
+
"rewards/chosen": 0.14953918890519577,
|
| 143 |
+
"rewards/margins": 0.10617925031016573,
|
| 144 |
+
"rewards/rejected": 0.04335993859503004,
|
| 145 |
+
"step": 80
|
| 146 |
+
},
|
| 147 |
+
{
|
| 148 |
+
"epoch": 2.0,
|
| 149 |
+
"grad_norm": 1.212666630744934,
|
| 150 |
+
"kl": 19.860130310058594,
|
| 151 |
+
"learning_rate": 4.849231551964771e-06,
|
| 152 |
+
"logits/chosen": -6533466.514285714,
|
| 153 |
+
"logits/rejected": -6697638.4,
|
| 154 |
+
"logps/chosen": -13.433639090401785,
|
| 155 |
+
"logps/rejected": -18.823931206597223,
|
| 156 |
+
"loss": 0.4773,
|
| 157 |
+
"rewards/chosen": 0.25568812234061106,
|
| 158 |
+
"rewards/margins": 0.21222362593998986,
|
| 159 |
+
"rewards/rejected": 0.0434644964006212,
|
| 160 |
+
"step": 90
|
| 161 |
+
},
|
| 162 |
+
{
|
| 163 |
+
"epoch": 2.2222222222222223,
|
| 164 |
+
"grad_norm": 1.1951372623443604,
|
| 165 |
+
"kl": 18.933120727539062,
|
| 166 |
+
"learning_rate": 4.775907352415367e-06,
|
| 167 |
+
"logits/chosen": -5909972.053333334,
|
| 168 |
+
"logits/rejected": -7999505.3176470585,
|
| 169 |
+
"logps/chosen": -13.5289892578125,
|
| 170 |
+
"logps/rejected": -18.93715676700368,
|
| 171 |
+
"loss": 0.4593,
|
| 172 |
+
"rewards/chosen": 0.3267870839436849,
|
| 173 |
+
"rewards/margins": 0.3485112210816028,
|
| 174 |
+
"rewards/rejected": -0.02172413713791791,
|
| 175 |
+
"step": 100
|
| 176 |
+
},
|
| 177 |
+
{
|
| 178 |
+
"epoch": 2.2222222222222223,
|
| 179 |
+
"eval_logits/chosen": -6310909.257142857,
|
| 180 |
+
"eval_logits/rejected": -7175433.955555555,
|
| 181 |
+
"eval_logps/chosen": -14.033848353794642,
|
| 182 |
+
"eval_logps/rejected": -19.490941026475696,
|
| 183 |
+
"eval_loss": 0.46696051955223083,
|
| 184 |
+
"eval_rewards/chosen": 0.3125021525791713,
|
| 185 |
+
"eval_rewards/margins": 0.29760706311180474,
|
| 186 |
+
"eval_rewards/rejected": 0.014895089467366536,
|
| 187 |
+
"eval_runtime": 17.481,
|
| 188 |
+
"eval_samples_per_second": 4.576,
|
| 189 |
+
"eval_steps_per_second": 2.288,
|
| 190 |
+
"kl": 7.871938228607178,
|
| 191 |
+
"step": 100
|
| 192 |
+
},
|
| 193 |
+
{
|
| 194 |
+
"epoch": 2.4444444444444446,
|
| 195 |
+
"grad_norm": 1.7072739601135254,
|
| 196 |
+
"kl": 5.876051902770996,
|
| 197 |
+
"learning_rate": 4.688895578255228e-06,
|
| 198 |
+
"logits/chosen": -4674539.2,
|
| 199 |
+
"logits/rejected": -7180268.0,
|
| 200 |
+
"logps/chosen": -11.73232650756836,
|
| 201 |
+
"logps/rejected": -21.302752685546874,
|
| 202 |
+
"loss": 0.4337,
|
| 203 |
+
"rewards/chosen": 0.4022516250610352,
|
| 204 |
+
"rewards/margins": 0.5421271562576294,
|
| 205 |
+
"rewards/rejected": -0.13987553119659424,
|
| 206 |
+
"step": 110
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"epoch": 2.6666666666666665,
|
| 210 |
+
"grad_norm": 1.409259557723999,
|
| 211 |
+
"kl": 15.80307674407959,
|
| 212 |
+
"learning_rate": 4.588719528532342e-06,
|
| 213 |
+
"logits/chosen": -5999085.176470588,
|
| 214 |
+
"logits/rejected": -6937394.773333333,
|
| 215 |
+
"logps/chosen": -15.016993623621323,
|
| 216 |
+
"logps/rejected": -23.880426432291667,
|
| 217 |
+
"loss": 0.4161,
|
| 218 |
+
"rewards/chosen": 0.3759063271915211,
|
| 219 |
+
"rewards/margins": 0.6798269204532399,
|
| 220 |
+
"rewards/rejected": -0.3039205932617188,
|
| 221 |
+
"step": 120
|
| 222 |
+
},
|
| 223 |
+
{
|
| 224 |
+
"epoch": 2.888888888888889,
|
| 225 |
+
"grad_norm": 1.5020368099212646,
|
| 226 |
+
"kl": 4.061548233032227,
|
| 227 |
+
"learning_rate": 4.475981673796899e-06,
|
| 228 |
+
"logits/chosen": -5263017.316455696,
|
| 229 |
+
"logits/rejected": -6793347.160493827,
|
| 230 |
+
"logps/chosen": -13.852985623516615,
|
| 231 |
+
"logps/rejected": -24.90943287037037,
|
| 232 |
+
"loss": 0.4033,
|
| 233 |
+
"rewards/chosen": 0.3698826560491248,
|
| 234 |
+
"rewards/margins": 0.8227180216271051,
|
| 235 |
+
"rewards/rejected": -0.4528353655779803,
|
| 236 |
+
"step": 130
|
| 237 |
+
},
|
| 238 |
+
{
|
| 239 |
+
"epoch": 3.111111111111111,
|
| 240 |
+
"grad_norm": 1.3534297943115234,
|
| 241 |
+
"kl": 10.132991790771484,
|
| 242 |
+
"learning_rate": 4.351360032772512e-06,
|
| 243 |
+
"logits/chosen": -5211370.06741573,
|
| 244 |
+
"logits/rejected": -6893736.563380281,
|
| 245 |
+
"logps/chosen": -13.59007417485955,
|
| 246 |
+
"logps/rejected": -28.269847601232396,
|
| 247 |
+
"loss": 0.3634,
|
| 248 |
+
"rewards/chosen": 0.4416037141607049,
|
| 249 |
+
"rewards/margins": 1.272281283912713,
|
| 250 |
+
"rewards/rejected": -0.8306775697520081,
|
| 251 |
+
"step": 140
|
| 252 |
+
},
|
| 253 |
+
{
|
| 254 |
+
"epoch": 3.3333333333333335,
|
| 255 |
+
"grad_norm": 1.6547423601150513,
|
| 256 |
+
"kl": 4.284663677215576,
|
| 257 |
+
"learning_rate": 4.215604094671835e-06,
|
| 258 |
+
"logits/chosen": -4668295.724137931,
|
| 259 |
+
"logits/rejected": -6887783.890410959,
|
| 260 |
+
"logps/chosen": -13.891624012212644,
|
| 261 |
+
"logps/rejected": -28.655340325342465,
|
| 262 |
+
"loss": 0.3701,
|
| 263 |
+
"rewards/chosen": 0.3276262392942933,
|
| 264 |
+
"rewards/margins": 1.2273214397571572,
|
| 265 |
+
"rewards/rejected": -0.8996952004628639,
|
| 266 |
+
"step": 150
|
| 267 |
+
},
|
| 268 |
+
{
|
| 269 |
+
"epoch": 3.3333333333333335,
|
| 270 |
+
"eval_logits/chosen": -5641927.314285714,
|
| 271 |
+
"eval_logits/rejected": -6731665.777777778,
|
| 272 |
+
"eval_logps/chosen": -14.361015101841518,
|
| 273 |
+
"eval_logps/rejected": -29.413007269965277,
|
| 274 |
+
"eval_loss": 0.36061567068099976,
|
| 275 |
+
"eval_rewards/chosen": 0.27978545597621374,
|
| 276 |
+
"eval_rewards/margins": 1.257096823434981,
|
| 277 |
+
"eval_rewards/rejected": -0.9773113674587673,
|
| 278 |
+
"eval_runtime": 17.4799,
|
| 279 |
+
"eval_samples_per_second": 4.577,
|
| 280 |
+
"eval_steps_per_second": 2.288,
|
| 281 |
+
"kl": 0.0,
|
| 282 |
+
"step": 150
|
| 283 |
+
},
|
| 284 |
+
{
|
| 285 |
+
"epoch": 3.5555555555555554,
|
| 286 |
+
"grad_norm": 1.6560704708099365,
|
| 287 |
+
"kl": 0.0,
|
| 288 |
+
"learning_rate": 4.069530311680247e-06,
|
| 289 |
+
"logits/chosen": -4197209.6,
|
| 290 |
+
"logits/rejected": -6773993.411764706,
|
| 291 |
+
"logps/chosen": -9.26181640625,
|
| 292 |
+
"logps/rejected": -33.01460248161764,
|
| 293 |
+
"loss": 0.3012,
|
| 294 |
+
"rewards/chosen": 0.6008858235677084,
|
| 295 |
+
"rewards/margins": 1.9487138695810355,
|
| 296 |
+
"rewards/rejected": -1.3478280460133272,
|
| 297 |
+
"step": 160
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"epoch": 3.7777777777777777,
|
| 301 |
+
"grad_norm": 4.158433437347412,
|
| 302 |
+
"kl": 6.74577522277832,
|
| 303 |
+
"learning_rate": 3.914017188716347e-06,
|
| 304 |
+
"logits/chosen": -4469396.0,
|
| 305 |
+
"logits/rejected": -6473352.0,
|
| 306 |
+
"logps/chosen": -14.897393798828125,
|
| 307 |
+
"logps/rejected": -32.791632080078124,
|
| 308 |
+
"loss": 0.3442,
|
| 309 |
+
"rewards/chosen": 0.3283118724822998,
|
| 310 |
+
"rewards/margins": 1.6118808269500733,
|
| 311 |
+
"rewards/rejected": -1.2835689544677735,
|
| 312 |
+
"step": 170
|
| 313 |
+
},
|
| 314 |
+
{
|
| 315 |
+
"epoch": 4.0,
|
| 316 |
+
"grad_norm": 1.2808359861373901,
|
| 317 |
+
"kl": 6.034703254699707,
|
| 318 |
+
"learning_rate": 3.7500000000000005e-06,
|
| 319 |
+
"logits/chosen": -5142258.4,
|
| 320 |
+
"logits/rejected": -6658021.6,
|
| 321 |
+
"logps/chosen": -12.736177062988281,
|
| 322 |
+
"logps/rejected": -38.654931640625,
|
| 323 |
+
"loss": 0.2877,
|
| 324 |
+
"rewards/chosen": 0.5321792602539063,
|
| 325 |
+
"rewards/margins": 2.298000717163086,
|
| 326 |
+
"rewards/rejected": -1.7658214569091797,
|
| 327 |
+
"step": 180
|
| 328 |
+
},
|
| 329 |
+
{
|
| 330 |
+
"epoch": 4.222222222222222,
|
| 331 |
+
"grad_norm": 4.354324817657471,
|
| 332 |
+
"kl": 9.844001770019531,
|
| 333 |
+
"learning_rate": 3.578465164203134e-06,
|
| 334 |
+
"logits/chosen": -4434924.873563218,
|
| 335 |
+
"logits/rejected": -6367553.7534246575,
|
| 336 |
+
"logps/chosen": -10.876662681842673,
|
| 337 |
+
"logps/rejected": -37.46076359160959,
|
| 338 |
+
"loss": 0.2869,
|
| 339 |
+
"rewards/chosen": 0.6059916704550556,
|
| 340 |
+
"rewards/margins": 2.4776278390525515,
|
| 341 |
+
"rewards/rejected": -1.8716361685974958,
|
| 342 |
+
"step": 190
|
| 343 |
+
},
|
| 344 |
+
{
|
| 345 |
+
"epoch": 4.444444444444445,
|
| 346 |
+
"grad_norm": 1.9371322393417358,
|
| 347 |
+
"kl": 0.0,
|
| 348 |
+
"learning_rate": 3.400444312011776e-06,
|
| 349 |
+
"logits/chosen": -5190637.948717949,
|
| 350 |
+
"logits/rejected": -6783854.048780488,
|
| 351 |
+
"logps/chosen": -11.751712114383013,
|
| 352 |
+
"logps/rejected": -40.603593035442074,
|
| 353 |
+
"loss": 0.281,
|
| 354 |
+
"rewards/chosen": 0.5476302122458433,
|
| 355 |
+
"rewards/margins": 2.449139008751655,
|
| 356 |
+
"rewards/rejected": -1.9015087965058117,
|
| 357 |
+
"step": 200
|
| 358 |
+
},
|
| 359 |
+
{
|
| 360 |
+
"epoch": 4.444444444444445,
|
| 361 |
+
"eval_logits/chosen": -5451904.914285715,
|
| 362 |
+
"eval_logits/rejected": -6268727.466666667,
|
| 363 |
+
"eval_logps/chosen": -15.457679966517857,
|
| 364 |
+
"eval_logps/rejected": -40.028545464409724,
|
| 365 |
+
"eval_loss": 0.3004380762577057,
|
| 366 |
+
"eval_rewards/chosen": 0.17011917659214565,
|
| 367 |
+
"eval_rewards/margins": 2.2089846686711385,
|
| 368 |
+
"eval_rewards/rejected": -2.038865492078993,
|
| 369 |
+
"eval_runtime": 17.4949,
|
| 370 |
+
"eval_samples_per_second": 4.573,
|
| 371 |
+
"eval_steps_per_second": 2.286,
|
| 372 |
+
"kl": 0.0,
|
| 373 |
+
"step": 200
|
| 374 |
+
},
|
| 375 |
+
{
|
| 376 |
+
"epoch": 4.666666666666667,
|
| 377 |
+
"grad_norm": 2.124127149581909,
|
| 378 |
+
"kl": 0.34170961380004883,
|
| 379 |
+
"learning_rate": 3.217008081777726e-06,
|
| 380 |
+
"logits/chosen": -5209803.317073171,
|
| 381 |
+
"logits/rejected": -6119242.666666667,
|
| 382 |
+
"logps/chosen": -12.961345393483231,
|
| 383 |
+
"logps/rejected": -43.778980744190704,
|
| 384 |
+
"loss": 0.2759,
|
| 385 |
+
"rewards/chosen": 0.4929830969833746,
|
| 386 |
+
"rewards/margins": 2.855773718227365,
|
| 387 |
+
"rewards/rejected": -2.3627906212439904,
|
| 388 |
+
"step": 210
|
| 389 |
+
},
|
| 390 |
+
{
|
| 391 |
+
"epoch": 4.888888888888889,
|
| 392 |
+
"grad_norm": 5.048134803771973,
|
| 393 |
+
"kl": 0.299468994140625,
|
| 394 |
+
"learning_rate": 3.0292596805735275e-06,
|
| 395 |
+
"logits/chosen": -3677977.6,
|
| 396 |
+
"logits/rejected": -7151626.4,
|
| 397 |
+
"logps/chosen": -11.91136474609375,
|
| 398 |
+
"logps/rejected": -44.98524780273438,
|
| 399 |
+
"loss": 0.2552,
|
| 400 |
+
"rewards/chosen": 0.5031857490539551,
|
| 401 |
+
"rewards/margins": 3.0173224449157714,
|
| 402 |
+
"rewards/rejected": -2.5141366958618163,
|
| 403 |
+
"step": 220
|
| 404 |
+
},
|
| 405 |
+
{
|
| 406 |
+
"epoch": 5.111111111111111,
|
| 407 |
+
"grad_norm": 1.2531408071517944,
|
| 408 |
+
"kl": 0.0,
|
| 409 |
+
"learning_rate": 2.8383282493753282e-06,
|
| 410 |
+
"logits/chosen": -4280043.220779221,
|
| 411 |
+
"logits/rejected": -6925918.072289157,
|
| 412 |
+
"logps/chosen": -12.665734229149756,
|
| 413 |
+
"logps/rejected": -43.957440112010545,
|
| 414 |
+
"loss": 0.2576,
|
| 415 |
+
"rewards/chosen": 0.7539182390485492,
|
| 416 |
+
"rewards/margins": 3.1649977377072465,
|
| 417 |
+
"rewards/rejected": -2.4110794986586974,
|
| 418 |
+
"step": 230
|
| 419 |
+
},
|
| 420 |
+
{
|
| 421 |
+
"epoch": 5.333333333333333,
|
| 422 |
+
"grad_norm": 5.549612998962402,
|
| 423 |
+
"kl": 0.0,
|
| 424 |
+
"learning_rate": 2.6453620722761897e-06,
|
| 425 |
+
"logits/chosen": -5263511.717647059,
|
| 426 |
+
"logits/rejected": -5916718.08,
|
| 427 |
+
"logps/chosen": -9.854262408088236,
|
| 428 |
+
"logps/rejected": -46.6882421875,
|
| 429 |
+
"loss": 0.2567,
|
| 430 |
+
"rewards/chosen": 0.6036793877096737,
|
| 431 |
+
"rewards/margins": 3.234525131524778,
|
| 432 |
+
"rewards/rejected": -2.630845743815104,
|
| 433 |
+
"step": 240
|
| 434 |
+
},
|
| 435 |
+
{
|
| 436 |
+
"epoch": 5.555555555555555,
|
| 437 |
+
"grad_norm": 1.6388943195343018,
|
| 438 |
+
"kl": 0.0,
|
| 439 |
+
"learning_rate": 2.4515216705704396e-06,
|
| 440 |
+
"logits/chosen": -4496943.157894737,
|
| 441 |
+
"logits/rejected": -6177542.857142857,
|
| 442 |
+
"logps/chosen": -10.795360364412007,
|
| 443 |
+
"logps/rejected": -54.87718563988095,
|
| 444 |
+
"loss": 0.2051,
|
| 445 |
+
"rewards/chosen": 0.5903302242881373,
|
| 446 |
+
"rewards/margins": 4.18927874839993,
|
| 447 |
+
"rewards/rejected": -3.598948524111793,
|
| 448 |
+
"step": 250
|
| 449 |
+
},
|
| 450 |
+
{
|
| 451 |
+
"epoch": 5.555555555555555,
|
| 452 |
+
"eval_logits/chosen": -5351382.857142857,
|
| 453 |
+
"eval_logits/rejected": -5877686.755555555,
|
| 454 |
+
"eval_logps/chosen": -15.197433035714285,
|
| 455 |
+
"eval_logps/rejected": -48.05066189236111,
|
| 456 |
+
"eval_loss": 0.27403125166893005,
|
| 457 |
+
"eval_rewards/chosen": 0.19614369528634207,
|
| 458 |
+
"eval_rewards/margins": 3.0372208640688942,
|
| 459 |
+
"eval_rewards/rejected": -2.8410771687825522,
|
| 460 |
+
"eval_runtime": 17.5075,
|
| 461 |
+
"eval_samples_per_second": 4.569,
|
| 462 |
+
"eval_steps_per_second": 2.285,
|
| 463 |
+
"kl": 0.0,
|
| 464 |
+
"step": 250
|
| 465 |
+
},
|
| 466 |
+
{
|
| 467 |
+
"epoch": 5.777777777777778,
|
| 468 |
+
"grad_norm": 2.7054665088653564,
|
| 469 |
+
"kl": 1.0151314735412598,
|
| 470 |
+
"learning_rate": 2.2579728232420524e-06,
|
| 471 |
+
"logits/chosen": -4953812.819277109,
|
| 472 |
+
"logits/rejected": -6552532.779220779,
|
| 473 |
+
"logps/chosen": -11.884703119117093,
|
| 474 |
+
"logps/rejected": -51.183222783076296,
|
| 475 |
+
"loss": 0.2612,
|
| 476 |
+
"rewards/chosen": 0.3994415467043957,
|
| 477 |
+
"rewards/margins": 3.453126722612472,
|
| 478 |
+
"rewards/rejected": -3.0536851759080763,
|
| 479 |
+
"step": 260
|
| 480 |
+
},
|
| 481 |
+
{
|
| 482 |
+
"epoch": 6.0,
|
| 483 |
+
"grad_norm": 0.8331411480903625,
|
| 484 |
+
"kl": 0.0,
|
| 485 |
+
"learning_rate": 2.0658795558326745e-06,
|
| 486 |
+
"logits/chosen": -4686625.951219512,
|
| 487 |
+
"logits/rejected": -7812360.205128205,
|
| 488 |
+
"logps/chosen": -9.980894507431403,
|
| 489 |
+
"logps/rejected": -53.11843950320513,
|
| 490 |
+
"loss": 0.221,
|
| 491 |
+
"rewards/chosen": 0.8428270293445121,
|
| 492 |
+
"rewards/margins": 4.1761818814829335,
|
| 493 |
+
"rewards/rejected": -3.3333548521384215,
|
| 494 |
+
"step": 270
|
| 495 |
+
},
|
| 496 |
+
{
|
| 497 |
+
"epoch": 6.222222222222222,
|
| 498 |
+
"grad_norm": 1.117002248764038,
|
| 499 |
+
"kl": 15.995762825012207,
|
| 500 |
+
"learning_rate": 1.876397139855047e-06,
|
| 501 |
+
"logits/chosen": -4876782.12987013,
|
| 502 |
+
"logits/rejected": -6101197.879518072,
|
| 503 |
+
"logps/chosen": -11.705315131645698,
|
| 504 |
+
"logps/rejected": -58.6199171686747,
|
| 505 |
+
"loss": 0.2091,
|
| 506 |
+
"rewards/chosen": 0.6778353901652546,
|
| 507 |
+
"rewards/margins": 4.467821138823952,
|
| 508 |
+
"rewards/rejected": -3.7899857486586974,
|
| 509 |
+
"step": 280
|
| 510 |
+
},
|
| 511 |
+
{
|
| 512 |
+
"epoch": 6.444444444444445,
|
| 513 |
+
"grad_norm": 6.495964527130127,
|
| 514 |
+
"kl": 4.720248699188232,
|
| 515 |
+
"learning_rate": 1.6906651448541977e-06,
|
| 516 |
+
"logits/chosen": -3970153.0256410255,
|
| 517 |
+
"logits/rejected": -7171078.243902439,
|
| 518 |
+
"logps/chosen": -6.301290658804087,
|
| 519 |
+
"logps/rejected": -57.83571122332317,
|
| 520 |
+
"loss": 0.1688,
|
| 521 |
+
"rewards/chosen": 1.1230317140236878,
|
| 522 |
+
"rewards/margins": 4.893083995845334,
|
| 523 |
+
"rewards/rejected": -3.770052281821646,
|
| 524 |
+
"step": 290
|
| 525 |
+
},
|
| 526 |
+
{
|
| 527 |
+
"epoch": 6.666666666666667,
|
| 528 |
+
"grad_norm": 0.7937259674072266,
|
| 529 |
+
"kl": 0.0,
|
| 530 |
+
"learning_rate": 1.509800584902108e-06,
|
| 531 |
+
"logits/chosen": -5307403.341772152,
|
| 532 |
+
"logits/rejected": -6794199.703703703,
|
| 533 |
+
"logps/chosen": -16.510799359671676,
|
| 534 |
+
"logps/rejected": -55.48344666280864,
|
| 535 |
+
"loss": 0.2724,
|
| 536 |
+
"rewards/chosen": 0.0943850988074194,
|
| 537 |
+
"rewards/margins": 3.622630812783114,
|
| 538 |
+
"rewards/rejected": -3.5282457139756946,
|
| 539 |
+
"step": 300
|
| 540 |
+
},
|
| 541 |
+
{
|
| 542 |
+
"epoch": 6.666666666666667,
|
| 543 |
+
"eval_logits/chosen": -5272125.257142857,
|
| 544 |
+
"eval_logits/rejected": -5524427.377777778,
|
| 545 |
+
"eval_logps/chosen": -16.101872907366072,
|
| 546 |
+
"eval_logps/rejected": -56.35105251736111,
|
| 547 |
+
"eval_loss": 0.2627750039100647,
|
| 548 |
+
"eval_rewards/chosen": 0.10569986615862165,
|
| 549 |
+
"eval_rewards/margins": 3.776815656631712,
|
| 550 |
+
"eval_rewards/rejected": -3.6711157904730904,
|
| 551 |
+
"eval_runtime": 17.4909,
|
| 552 |
+
"eval_samples_per_second": 4.574,
|
| 553 |
+
"eval_steps_per_second": 2.287,
|
| 554 |
+
"kl": 0.0,
|
| 555 |
+
"step": 300
|
| 556 |
+
},
|
| 557 |
+
{
|
| 558 |
+
"epoch": 6.888888888888889,
|
| 559 |
+
"grad_norm": 1.1168758869171143,
|
| 560 |
+
"kl": 0.0,
|
| 561 |
+
"learning_rate": 1.3348912007436538e-06,
|
| 562 |
+
"logits/chosen": -3931129.379310345,
|
| 563 |
+
"logits/rejected": -6447021.150684931,
|
| 564 |
+
"logps/chosen": -6.620521282327586,
|
| 565 |
+
"logps/rejected": -63.137895976027394,
|
| 566 |
+
"loss": 0.1898,
|
| 567 |
+
"rewards/chosen": 1.0009118441877694,
|
| 568 |
+
"rewards/margins": 5.434586277647744,
|
| 569 |
+
"rewards/rejected": -4.4336744334599745,
|
| 570 |
+
"step": 310
|
| 571 |
+
},
|
| 572 |
+
{
|
| 573 |
+
"epoch": 7.111111111111111,
|
| 574 |
+
"grad_norm": 3.610934019088745,
|
| 575 |
+
"kl": 0.0,
|
| 576 |
+
"learning_rate": 1.1669889179957725e-06,
|
| 577 |
+
"logits/chosen": -4449015.636363637,
|
| 578 |
+
"logits/rejected": -6512209.333333333,
|
| 579 |
+
"logps/chosen": -9.530960083007812,
|
| 580 |
+
"logps/rejected": -60.69569227430556,
|
| 581 |
+
"loss": 0.2499,
|
| 582 |
+
"rewards/chosen": 0.6914540204134855,
|
| 583 |
+
"rewards/margins": 4.681384067342739,
|
| 584 |
+
"rewards/rejected": -3.9899300469292536,
|
| 585 |
+
"step": 320
|
| 586 |
+
},
|
| 587 |
+
{
|
| 588 |
+
"epoch": 7.333333333333333,
|
| 589 |
+
"grad_norm": 5.70353889465332,
|
| 590 |
+
"kl": 7.343207359313965,
|
| 591 |
+
"learning_rate": 1.0071035207430352e-06,
|
| 592 |
+
"logits/chosen": -4714173.5696202535,
|
| 593 |
+
"logits/rejected": -5926990.222222222,
|
| 594 |
+
"logps/chosen": -9.35969717291337,
|
| 595 |
+
"logps/rejected": -61.245768229166664,
|
| 596 |
+
"loss": 0.1972,
|
| 597 |
+
"rewards/chosen": 0.7530057882960839,
|
| 598 |
+
"rewards/margins": 4.901618858560358,
|
| 599 |
+
"rewards/rejected": -4.1486130702642745,
|
| 600 |
+
"step": 330
|
| 601 |
+
},
|
| 602 |
+
{
|
| 603 |
+
"epoch": 7.555555555555555,
|
| 604 |
+
"grad_norm": 0.47463521361351013,
|
| 605 |
+
"kl": 0.0,
|
| 606 |
+
"learning_rate": 8.561965785773413e-07,
|
| 607 |
+
"logits/chosen": -3589752.0,
|
| 608 |
+
"logits/rejected": -6144573.333333333,
|
| 609 |
+
"logps/chosen": -10.513381264426492,
|
| 610 |
+
"logps/rejected": -67.19074164496527,
|
| 611 |
+
"loss": 0.2194,
|
| 612 |
+
"rewards/chosen": 0.5275679935108532,
|
| 613 |
+
"rewards/margins": 5.252333771098744,
|
| 614 |
+
"rewards/rejected": -4.724765777587891,
|
| 615 |
+
"step": 340
|
| 616 |
+
},
|
| 617 |
+
{
|
| 618 |
+
"epoch": 7.777777777777778,
|
| 619 |
+
"grad_norm": 0.7246691584587097,
|
| 620 |
+
"kl": 9.73458194732666,
|
| 621 |
+
"learning_rate": 7.151756636052529e-07,
|
| 622 |
+
"logits/chosen": -4579355.654320988,
|
| 623 |
+
"logits/rejected": -5925367.898734177,
|
| 624 |
+
"logps/chosen": -13.632544246720679,
|
| 625 |
+
"logps/rejected": -63.80702383306962,
|
| 626 |
+
"loss": 0.2237,
|
| 627 |
+
"rewards/chosen": 0.650694199550299,
|
| 628 |
+
"rewards/margins": 4.9783826822935735,
|
| 629 |
+
"rewards/rejected": -4.327688482743275,
|
| 630 |
+
"step": 350
|
| 631 |
+
},
|
| 632 |
+
{
|
| 633 |
+
"epoch": 7.777777777777778,
|
| 634 |
+
"eval_logits/chosen": -5216298.057142857,
|
| 635 |
+
"eval_logits/rejected": -5306349.866666666,
|
| 636 |
+
"eval_logps/chosen": -16.67706298828125,
|
| 637 |
+
"eval_logps/rejected": -61.12718098958333,
|
| 638 |
+
"eval_loss": 0.2568877041339874,
|
| 639 |
+
"eval_rewards/chosen": 0.0481806891305106,
|
| 640 |
+
"eval_rewards/margins": 4.196909462459503,
|
| 641 |
+
"eval_rewards/rejected": -4.148728773328993,
|
| 642 |
+
"eval_runtime": 17.4735,
|
| 643 |
+
"eval_samples_per_second": 4.578,
|
| 644 |
+
"eval_steps_per_second": 2.289,
|
| 645 |
+
"kl": 0.0,
|
| 646 |
+
"step": 350
|
| 647 |
+
},
|
| 648 |
+
{
|
| 649 |
+
"epoch": 8.0,
|
| 650 |
+
"grad_norm": 1.7622766494750977,
|
| 651 |
+
"kl": 0.0,
|
| 652 |
+
"learning_rate": 5.848888922025553e-07,
|
| 653 |
+
"logits/chosen": -4720167.890410959,
|
| 654 |
+
"logits/rejected": -5690909.425287357,
|
| 655 |
+
"logps/chosen": -7.494754320954623,
|
| 656 |
+
"logps/rejected": -63.13269463900862,
|
| 657 |
+
"loss": 0.1865,
|
| 658 |
+
"rewards/chosen": 0.9349731027263485,
|
| 659 |
+
"rewards/margins": 5.348946505429294,
|
| 660 |
+
"rewards/rejected": -4.413973402702945,
|
| 661 |
+
"step": 360
|
| 662 |
+
},
|
| 663 |
+
{
|
| 664 |
+
"epoch": 8.222222222222221,
|
| 665 |
+
"grad_norm": 1.9263681173324585,
|
| 666 |
+
"kl": 0.0,
|
| 667 |
+
"learning_rate": 4.661198243425813e-07,
|
| 668 |
+
"logits/chosen": -4662120.0,
|
| 669 |
+
"logits/rejected": -6811490.4,
|
| 670 |
+
"logps/chosen": -11.226513671875,
|
| 671 |
+
"logps/rejected": -60.887017822265626,
|
| 672 |
+
"loss": 0.2197,
|
| 673 |
+
"rewards/chosen": 0.5319944381713867,
|
| 674 |
+
"rewards/margins": 4.648596000671386,
|
| 675 |
+
"rewards/rejected": -4.1166015625,
|
| 676 |
+
"step": 370
|
| 677 |
+
},
|
| 678 |
+
{
|
| 679 |
+
"epoch": 8.444444444444445,
|
| 680 |
+
"grad_norm": 2.094463348388672,
|
| 681 |
+
"kl": 1.3769874572753906,
|
| 682 |
+
"learning_rate": 3.595827511743341e-07,
|
| 683 |
+
"logits/chosen": -4225483.377777778,
|
| 684 |
+
"logits/rejected": -6771605.485714286,
|
| 685 |
+
"logps/chosen": -10.853142632378471,
|
| 686 |
+
"logps/rejected": -65.53819754464286,
|
| 687 |
+
"loss": 0.232,
|
| 688 |
+
"rewards/chosen": 0.6328274197048611,
|
| 689 |
+
"rewards/margins": 5.1289795890687,
|
| 690 |
+
"rewards/rejected": -4.496152169363839,
|
| 691 |
+
"step": 380
|
| 692 |
+
},
|
| 693 |
+
{
|
| 694 |
+
"epoch": 8.666666666666666,
|
| 695 |
+
"grad_norm": 1.4835810661315918,
|
| 696 |
+
"kl": 0.0,
|
| 697 |
+
"learning_rate": 2.6591839919146963e-07,
|
| 698 |
+
"logits/chosen": -4116534.222222222,
|
| 699 |
+
"logits/rejected": -5742267.636363637,
|
| 700 |
+
"logps/chosen": -7.460245768229167,
|
| 701 |
+
"logps/rejected": -67.84841086647727,
|
| 702 |
+
"loss": 0.1502,
|
| 703 |
+
"rewards/chosen": 1.145952648586697,
|
| 704 |
+
"rewards/margins": 6.005312389797634,
|
| 705 |
+
"rewards/rejected": -4.8593597412109375,
|
| 706 |
+
"step": 390
|
| 707 |
+
},
|
| 708 |
+
{
|
| 709 |
+
"epoch": 8.88888888888889,
|
| 710 |
+
"grad_norm": 0.8003972172737122,
|
| 711 |
+
"kl": 0.0,
|
| 712 |
+
"learning_rate": 1.8569007682777417e-07,
|
| 713 |
+
"logits/chosen": -4092200.459770115,
|
| 714 |
+
"logits/rejected": -6279503.780821918,
|
| 715 |
+
"logps/chosen": -10.580308234554598,
|
| 716 |
+
"logps/rejected": -65.71065657106165,
|
| 717 |
+
"loss": 0.2291,
|
| 718 |
+
"rewards/chosen": 0.5882207936254041,
|
| 719 |
+
"rewards/margins": 5.146064145642741,
|
| 720 |
+
"rewards/rejected": -4.557843352017337,
|
| 721 |
+
"step": 400
|
| 722 |
+
},
|
| 723 |
+
{
|
| 724 |
+
"epoch": 8.88888888888889,
|
| 725 |
+
"eval_logits/chosen": -5214656.914285715,
|
| 726 |
+
"eval_logits/rejected": -5268033.422222222,
|
| 727 |
+
"eval_logps/chosen": -16.733164760044644,
|
| 728 |
+
"eval_logps/rejected": -62.435932074652776,
|
| 729 |
+
"eval_loss": 0.254804790019989,
|
| 730 |
+
"eval_rewards/chosen": 0.042570604596819196,
|
| 731 |
+
"eval_rewards/margins": 4.32217536805168,
|
| 732 |
+
"eval_rewards/rejected": -4.279604763454861,
|
| 733 |
+
"eval_runtime": 17.4792,
|
| 734 |
+
"eval_samples_per_second": 4.577,
|
| 735 |
+
"eval_steps_per_second": 2.288,
|
| 736 |
+
"kl": 0.0,
|
| 737 |
+
"step": 400
|
| 738 |
+
},
|
| 739 |
+
{
|
| 740 |
+
"epoch": 9.11111111111111,
|
| 741 |
+
"grad_norm": 4.520019054412842,
|
| 742 |
+
"kl": 2.1843814849853516,
|
| 743 |
+
"learning_rate": 1.1938028665396172e-07,
|
| 744 |
+
"logits/chosen": -4234930.849315069,
|
| 745 |
+
"logits/rejected": -5565993.931034483,
|
| 746 |
+
"logps/chosen": -15.401979211258562,
|
| 747 |
+
"logps/rejected": -69.01152792744253,
|
| 748 |
+
"loss": 0.2192,
|
| 749 |
+
"rewards/chosen": 0.15699117477626018,
|
| 750 |
+
"rewards/margins": 5.1128171558287026,
|
| 751 |
+
"rewards/rejected": -4.955825981052443,
|
| 752 |
+
"step": 410
|
| 753 |
+
},
|
| 754 |
+
{
|
| 755 |
+
"epoch": 9.333333333333334,
|
| 756 |
+
"grad_norm": 0.8408989310264587,
|
| 757 |
+
"kl": 1.832779884338379,
|
| 758 |
+
"learning_rate": 6.738782355044048e-08,
|
| 759 |
+
"logits/chosen": -5072888.558139535,
|
| 760 |
+
"logits/rejected": -4679876.324324325,
|
| 761 |
+
"logps/chosen": -13.259429221929507,
|
| 762 |
+
"logps/rejected": -67.32681191934121,
|
| 763 |
+
"loss": 0.2468,
|
| 764 |
+
"rewards/chosen": 0.3328412743501885,
|
| 765 |
+
"rewards/margins": 4.958795567266151,
|
| 766 |
+
"rewards/rejected": -4.625954292915963,
|
| 767 |
+
"step": 420
|
| 768 |
+
},
|
| 769 |
+
{
|
| 770 |
+
"epoch": 9.555555555555555,
|
| 771 |
+
"grad_norm": 0.5843256711959839,
|
| 772 |
+
"kl": 0.0,
|
| 773 |
+
"learning_rate": 3.0025376307977474e-08,
|
| 774 |
+
"logits/chosen": -4235825.34939759,
|
| 775 |
+
"logits/rejected": -6855219.532467533,
|
| 776 |
+
"logps/chosen": -12.172502264919052,
|
| 777 |
+
"logps/rejected": -70.21881341314935,
|
| 778 |
+
"loss": 0.1936,
|
| 779 |
+
"rewards/chosen": 0.7041290007441877,
|
| 780 |
+
"rewards/margins": 5.708083603270568,
|
| 781 |
+
"rewards/rejected": -5.00395460252638,
|
| 782 |
+
"step": 430
|
| 783 |
+
},
|
| 784 |
+
{
|
| 785 |
+
"epoch": 9.777777777777779,
|
| 786 |
+
"grad_norm": 1.3815499544143677,
|
| 787 |
+
"kl": 0.0,
|
| 788 |
+
"learning_rate": 7.517647080519941e-09,
|
| 789 |
+
"logits/chosen": -4151235.011764706,
|
| 790 |
+
"logits/rejected": -7243480.746666667,
|
| 791 |
+
"logps/chosen": -5.613843132467831,
|
| 792 |
+
"logps/rejected": -63.06897135416666,
|
| 793 |
+
"loss": 0.1925,
|
| 794 |
+
"rewards/chosen": 1.112983793370864,
|
| 795 |
+
"rewards/margins": 5.333501371495864,
|
| 796 |
+
"rewards/rejected": -4.220517578125,
|
| 797 |
+
"step": 440
|
| 798 |
+
},
|
| 799 |
+
{
|
| 800 |
+
"epoch": 10.0,
|
| 801 |
+
"grad_norm": 0.7369117736816406,
|
| 802 |
+
"kl": 0.0,
|
| 803 |
+
"learning_rate": 0.0,
|
| 804 |
+
"logits/chosen": -4828750.702702703,
|
| 805 |
+
"logits/rejected": -5816776.930232558,
|
| 806 |
+
"logps/chosen": -6.564434979413007,
|
| 807 |
+
"logps/rejected": -65.85319767441861,
|
| 808 |
+
"loss": 0.1677,
|
| 809 |
+
"rewards/chosen": 1.0981009199812606,
|
| 810 |
+
"rewards/margins": 5.781813901449284,
|
| 811 |
+
"rewards/rejected": -4.683712981468023,
|
| 812 |
+
"step": 450
|
| 813 |
+
},
|
| 814 |
+
{
|
| 815 |
+
"epoch": 10.0,
|
| 816 |
+
"eval_logits/chosen": -5221032.228571429,
|
| 817 |
+
"eval_logits/rejected": -5284203.377777778,
|
| 818 |
+
"eval_logps/chosen": -16.849844796316965,
|
| 819 |
+
"eval_logps/rejected": -62.744411892361114,
|
| 820 |
+
"eval_loss": 0.2541462779045105,
|
| 821 |
+
"eval_rewards/chosen": 0.030902576446533204,
|
| 822 |
+
"eval_rewards/margins": 4.341355186038547,
|
| 823 |
+
"eval_rewards/rejected": -4.310452609592014,
|
| 824 |
+
"eval_runtime": 17.4751,
|
| 825 |
+
"eval_samples_per_second": 4.578,
|
| 826 |
+
"eval_steps_per_second": 2.289,
|
| 827 |
+
"kl": 0.0,
|
| 828 |
+
"step": 450
|
| 829 |
+
},
|
| 830 |
+
{
|
| 831 |
+
"epoch": 10.0,
|
| 832 |
+
"step": 450,
|
| 833 |
+
"total_flos": 3.939427351343923e+16,
|
| 834 |
+
"train_loss": 0.30883768532011246,
|
| 835 |
+
"train_runtime": 2497.4829,
|
| 836 |
+
"train_samples_per_second": 2.883,
|
| 837 |
+
"train_steps_per_second": 0.18
|
| 838 |
+
}
|
| 839 |
+
],
|
| 840 |
+
"logging_steps": 10,
|
| 841 |
+
"max_steps": 450,
|
| 842 |
+
"num_input_tokens_seen": 0,
|
| 843 |
+
"num_train_epochs": 10,
|
| 844 |
+
"save_steps": 50,
|
| 845 |
+
"stateful_callbacks": {
|
| 846 |
+
"TrainerControl": {
|
| 847 |
+
"args": {
|
| 848 |
+
"should_epoch_stop": false,
|
| 849 |
+
"should_evaluate": false,
|
| 850 |
+
"should_log": false,
|
| 851 |
+
"should_save": true,
|
| 852 |
+
"should_training_stop": true
|
| 853 |
+
},
|
| 854 |
+
"attributes": {}
|
| 855 |
+
}
|
| 856 |
+
},
|
| 857 |
+
"total_flos": 3.939427351343923e+16,
|
| 858 |
+
"train_batch_size": 2,
|
| 859 |
+
"trial_name": null,
|
| 860 |
+
"trial_params": null
|
| 861 |
+
}
|
training_eval_loss.png
ADDED
|
training_loss.png
ADDED
|
training_rewards_chosen.png
ADDED
|