Commit
•
6d0d94d
1
Parent(s):
e635f0f
Model save
Browse files- README.md +68 -0
- all_results.json +9 -0
- generation_config.json +12 -0
- train_results.json +9 -0
- trainer_state.json +826 -0
README.md
ADDED
|
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model: meta-llama/Meta-Llama-3-8B-Instruct
|
| 3 |
+
library_name: transformers
|
| 4 |
+
model_name: llama3-8b-dpo
|
| 5 |
+
tags:
|
| 6 |
+
- generated_from_trainer
|
| 7 |
+
- trl
|
| 8 |
+
- dpo
|
| 9 |
+
licence: license
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# Model Card for llama3-8b-dpo
|
| 13 |
+
|
| 14 |
+
This model is a fine-tuned version of [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct).
|
| 15 |
+
It has been trained using [TRL](https://github.com/huggingface/trl).
|
| 16 |
+
|
| 17 |
+
## Quick start
|
| 18 |
+
|
| 19 |
+
```python
|
| 20 |
+
from transformers import pipeline
|
| 21 |
+
|
| 22 |
+
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
|
| 23 |
+
generator = pipeline("text-generation", model="PeterJinGo/llama3-8b-dpo", device="cuda")
|
| 24 |
+
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
|
| 25 |
+
print(output["generated_text"])
|
| 26 |
+
```
|
| 27 |
+
|
| 28 |
+
## Training procedure
|
| 29 |
+
|
| 30 |
+
[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/uiuc-dmg/huggingface/runs/1ov1vz7b)
|
| 31 |
+
|
| 32 |
+
This model was trained with DPO, a method introduced in [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://huggingface.co/papers/2305.18290).
|
| 33 |
+
|
| 34 |
+
### Framework versions
|
| 35 |
+
|
| 36 |
+
- TRL: 0.12.0
|
| 37 |
+
- Transformers: 4.46.1
|
| 38 |
+
- Pytorch: 2.1.2+cu121
|
| 39 |
+
- Datasets: 3.1.0
|
| 40 |
+
- Tokenizers: 0.20.2
|
| 41 |
+
|
| 42 |
+
## Citations
|
| 43 |
+
|
| 44 |
+
Cite DPO as:
|
| 45 |
+
|
| 46 |
+
```bibtex
|
| 47 |
+
@inproceedings{rafailov2023direct,
|
| 48 |
+
title = {{Direct Preference Optimization: Your Language Model is Secretly a Reward Model}},
|
| 49 |
+
author = {Rafael Rafailov and Archit Sharma and Eric Mitchell and Christopher D. Manning and Stefano Ermon and Chelsea Finn},
|
| 50 |
+
year = 2023,
|
| 51 |
+
booktitle = {Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, New Orleans, LA, USA, December 10 - 16, 2023},
|
| 52 |
+
url = {http://papers.nips.cc/paper_files/paper/2023/hash/a85b405ed65c6477a4fe8302b5e06ce7-Abstract-Conference.html},
|
| 53 |
+
editor = {Alice Oh and Tristan Naumann and Amir Globerson and Kate Saenko and Moritz Hardt and Sergey Levine},
|
| 54 |
+
}
|
| 55 |
+
```
|
| 56 |
+
|
| 57 |
+
Cite TRL as:
|
| 58 |
+
|
| 59 |
+
```bibtex
|
| 60 |
+
@misc{vonwerra2022trl,
|
| 61 |
+
title = {{TRL: Transformer Reinforcement Learning}},
|
| 62 |
+
author = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec},
|
| 63 |
+
year = 2020,
|
| 64 |
+
journal = {GitHub repository},
|
| 65 |
+
publisher = {GitHub},
|
| 66 |
+
howpublished = {\url{https://github.com/huggingface/trl}}
|
| 67 |
+
}
|
| 68 |
+
```
|
all_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 1.0,
|
| 3 |
+
"total_flos": 0.0,
|
| 4 |
+
"train_loss": 1.1650768583289748,
|
| 5 |
+
"train_runtime": 3769.2146,
|
| 6 |
+
"train_samples": 61134,
|
| 7 |
+
"train_samples_per_second": 16.219,
|
| 8 |
+
"train_steps_per_second": 0.127
|
| 9 |
+
}
|
generation_config.json
ADDED
|
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token_id": 128000,
|
| 3 |
+
"do_sample": true,
|
| 4 |
+
"eos_token_id": [
|
| 5 |
+
128001,
|
| 6 |
+
128009
|
| 7 |
+
],
|
| 8 |
+
"max_length": 4096,
|
| 9 |
+
"temperature": 0.6,
|
| 10 |
+
"top_p": 0.9,
|
| 11 |
+
"transformers_version": "4.46.1"
|
| 12 |
+
}
|
train_results.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 1.0,
|
| 3 |
+
"total_flos": 0.0,
|
| 4 |
+
"train_loss": 1.1650768583289748,
|
| 5 |
+
"train_runtime": 3769.2146,
|
| 6 |
+
"train_samples": 61134,
|
| 7 |
+
"train_samples_per_second": 16.219,
|
| 8 |
+
"train_steps_per_second": 0.127
|
| 9 |
+
}
|
trainer_state.json
ADDED
|
@@ -0,0 +1,826 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": null,
|
| 3 |
+
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 1.0,
|
| 5 |
+
"eval_steps": 100,
|
| 6 |
+
"global_step": 478,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.0020920502092050207,
|
| 13 |
+
"grad_norm": 11.329232130912374,
|
| 14 |
+
"learning_rate": 1.0416666666666666e-08,
|
| 15 |
+
"logits/chosen": -0.28515625,
|
| 16 |
+
"logits/rejected": -0.34765625,
|
| 17 |
+
"logps/chosen": -372.0,
|
| 18 |
+
"logps/rejected": -440.0,
|
| 19 |
+
"loss": 1.3828,
|
| 20 |
+
"rewards/accuracies": 0.0,
|
| 21 |
+
"rewards/chosen": 0.0,
|
| 22 |
+
"rewards/margins": 0.0,
|
| 23 |
+
"rewards/rejected": 0.0,
|
| 24 |
+
"step": 1
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.02092050209205021,
|
| 28 |
+
"grad_norm": 10.768270205160954,
|
| 29 |
+
"learning_rate": 1.0416666666666667e-07,
|
| 30 |
+
"logits/chosen": -0.271484375,
|
| 31 |
+
"logits/rejected": -0.220703125,
|
| 32 |
+
"logps/chosen": -320.0,
|
| 33 |
+
"logps/rejected": -310.0,
|
| 34 |
+
"loss": 1.3853,
|
| 35 |
+
"rewards/accuracies": 0.2152777761220932,
|
| 36 |
+
"rewards/chosen": 0.000476837158203125,
|
| 37 |
+
"rewards/margins": 0.0002956390380859375,
|
| 38 |
+
"rewards/rejected": 0.0001850128173828125,
|
| 39 |
+
"step": 10
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.04184100418410042,
|
| 43 |
+
"grad_norm": 9.438876866649906,
|
| 44 |
+
"learning_rate": 2.0833333333333333e-07,
|
| 45 |
+
"logits/chosen": -0.2109375,
|
| 46 |
+
"logits/rejected": -0.189453125,
|
| 47 |
+
"logps/chosen": -342.0,
|
| 48 |
+
"logps/rejected": -366.0,
|
| 49 |
+
"loss": 1.3843,
|
| 50 |
+
"rewards/accuracies": 0.23749999701976776,
|
| 51 |
+
"rewards/chosen": -0.0003910064697265625,
|
| 52 |
+
"rewards/margins": -0.000263214111328125,
|
| 53 |
+
"rewards/rejected": -0.00012493133544921875,
|
| 54 |
+
"step": 20
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"epoch": 0.06276150627615062,
|
| 58 |
+
"grad_norm": 10.93659663876472,
|
| 59 |
+
"learning_rate": 3.1249999999999997e-07,
|
| 60 |
+
"logits/chosen": -0.2421875,
|
| 61 |
+
"logits/rejected": -0.2060546875,
|
| 62 |
+
"logps/chosen": -352.0,
|
| 63 |
+
"logps/rejected": -328.0,
|
| 64 |
+
"loss": 1.3813,
|
| 65 |
+
"rewards/accuracies": 0.2874999940395355,
|
| 66 |
+
"rewards/chosen": 0.00286865234375,
|
| 67 |
+
"rewards/margins": 0.000423431396484375,
|
| 68 |
+
"rewards/rejected": 0.00244140625,
|
| 69 |
+
"step": 30
|
| 70 |
+
},
|
| 71 |
+
{
|
| 72 |
+
"epoch": 0.08368200836820083,
|
| 73 |
+
"grad_norm": 9.676059177655553,
|
| 74 |
+
"learning_rate": 4.1666666666666667e-07,
|
| 75 |
+
"logits/chosen": -0.1845703125,
|
| 76 |
+
"logits/rejected": -0.1494140625,
|
| 77 |
+
"logps/chosen": -314.0,
|
| 78 |
+
"logps/rejected": -320.0,
|
| 79 |
+
"loss": 1.3725,
|
| 80 |
+
"rewards/accuracies": 0.606249988079071,
|
| 81 |
+
"rewards/chosen": 0.0135498046875,
|
| 82 |
+
"rewards/margins": 0.0135498046875,
|
| 83 |
+
"rewards/rejected": -3.0547380447387695e-06,
|
| 84 |
+
"step": 40
|
| 85 |
+
},
|
| 86 |
+
{
|
| 87 |
+
"epoch": 0.10460251046025104,
|
| 88 |
+
"grad_norm": 9.798546885731437,
|
| 89 |
+
"learning_rate": 4.999733114418725e-07,
|
| 90 |
+
"logits/chosen": -0.20703125,
|
| 91 |
+
"logits/rejected": -0.1689453125,
|
| 92 |
+
"logps/chosen": -362.0,
|
| 93 |
+
"logps/rejected": -364.0,
|
| 94 |
+
"loss": 1.3524,
|
| 95 |
+
"rewards/accuracies": 0.59375,
|
| 96 |
+
"rewards/chosen": 0.037841796875,
|
| 97 |
+
"rewards/margins": 0.033203125,
|
| 98 |
+
"rewards/rejected": 0.0047607421875,
|
| 99 |
+
"step": 50
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"epoch": 0.12552301255230125,
|
| 103 |
+
"grad_norm": 9.60865688194641,
|
| 104 |
+
"learning_rate": 4.990398100856366e-07,
|
| 105 |
+
"logits/chosen": -0.2041015625,
|
| 106 |
+
"logits/rejected": -0.19140625,
|
| 107 |
+
"logps/chosen": -320.0,
|
| 108 |
+
"logps/rejected": -366.0,
|
| 109 |
+
"loss": 1.3346,
|
| 110 |
+
"rewards/accuracies": 0.675000011920929,
|
| 111 |
+
"rewards/chosen": 0.056884765625,
|
| 112 |
+
"rewards/margins": 0.053466796875,
|
| 113 |
+
"rewards/rejected": 0.003448486328125,
|
| 114 |
+
"step": 60
|
| 115 |
+
},
|
| 116 |
+
{
|
| 117 |
+
"epoch": 0.14644351464435146,
|
| 118 |
+
"grad_norm": 9.988151072194361,
|
| 119 |
+
"learning_rate": 4.967775735898179e-07,
|
| 120 |
+
"logits/chosen": -0.06103515625,
|
| 121 |
+
"logits/rejected": -0.068359375,
|
| 122 |
+
"logps/chosen": -316.0,
|
| 123 |
+
"logps/rejected": -334.0,
|
| 124 |
+
"loss": 1.2953,
|
| 125 |
+
"rewards/accuracies": 0.699999988079071,
|
| 126 |
+
"rewards/chosen": 0.06787109375,
|
| 127 |
+
"rewards/margins": 0.11376953125,
|
| 128 |
+
"rewards/rejected": -0.0458984375,
|
| 129 |
+
"step": 70
|
| 130 |
+
},
|
| 131 |
+
{
|
| 132 |
+
"epoch": 0.16736401673640167,
|
| 133 |
+
"grad_norm": 12.035906986217713,
|
| 134 |
+
"learning_rate": 4.931986719649298e-07,
|
| 135 |
+
"logits/chosen": -0.1591796875,
|
| 136 |
+
"logits/rejected": -0.12255859375,
|
| 137 |
+
"logps/chosen": -384.0,
|
| 138 |
+
"logps/rejected": -340.0,
|
| 139 |
+
"loss": 1.272,
|
| 140 |
+
"rewards/accuracies": 0.6499999761581421,
|
| 141 |
+
"rewards/chosen": 0.050537109375,
|
| 142 |
+
"rewards/margins": 0.1533203125,
|
| 143 |
+
"rewards/rejected": -0.1025390625,
|
| 144 |
+
"step": 80
|
| 145 |
+
},
|
| 146 |
+
{
|
| 147 |
+
"epoch": 0.18828451882845187,
|
| 148 |
+
"grad_norm": 11.972506176579511,
|
| 149 |
+
"learning_rate": 4.883222001996351e-07,
|
| 150 |
+
"logits/chosen": -0.12060546875,
|
| 151 |
+
"logits/rejected": -0.11865234375,
|
| 152 |
+
"logps/chosen": -328.0,
|
| 153 |
+
"logps/rejected": -352.0,
|
| 154 |
+
"loss": 1.2271,
|
| 155 |
+
"rewards/accuracies": 0.731249988079071,
|
| 156 |
+
"rewards/chosen": -0.03955078125,
|
| 157 |
+
"rewards/margins": 0.22265625,
|
| 158 |
+
"rewards/rejected": -0.26171875,
|
| 159 |
+
"step": 90
|
| 160 |
+
},
|
| 161 |
+
{
|
| 162 |
+
"epoch": 0.20920502092050208,
|
| 163 |
+
"grad_norm": 11.425153872085387,
|
| 164 |
+
"learning_rate": 4.821741763807186e-07,
|
| 165 |
+
"logits/chosen": -0.2119140625,
|
| 166 |
+
"logits/rejected": -0.23046875,
|
| 167 |
+
"logps/chosen": -340.0,
|
| 168 |
+
"logps/rejected": -366.0,
|
| 169 |
+
"loss": 1.2223,
|
| 170 |
+
"rewards/accuracies": 0.6875,
|
| 171 |
+
"rewards/chosen": -0.1083984375,
|
| 172 |
+
"rewards/margins": 0.27734375,
|
| 173 |
+
"rewards/rejected": -0.38671875,
|
| 174 |
+
"step": 100
|
| 175 |
+
},
|
| 176 |
+
{
|
| 177 |
+
"epoch": 0.20920502092050208,
|
| 178 |
+
"eval_logits/chosen": -0.205078125,
|
| 179 |
+
"eval_logits/rejected": -0.150390625,
|
| 180 |
+
"eval_logps/chosen": -346.0,
|
| 181 |
+
"eval_logps/rejected": -390.0,
|
| 182 |
+
"eval_loss": 0.6163906455039978,
|
| 183 |
+
"eval_rewards/accuracies": 0.734375,
|
| 184 |
+
"eval_rewards/chosen": -0.1064453125,
|
| 185 |
+
"eval_rewards/margins": 0.318359375,
|
| 186 |
+
"eval_rewards/rejected": -0.42578125,
|
| 187 |
+
"eval_runtime": 44.8743,
|
| 188 |
+
"eval_samples_per_second": 44.569,
|
| 189 |
+
"eval_steps_per_second": 0.713,
|
| 190 |
+
"step": 100
|
| 191 |
+
},
|
| 192 |
+
{
|
| 193 |
+
"epoch": 0.2301255230125523,
|
| 194 |
+
"grad_norm": 17.93262654003255,
|
| 195 |
+
"learning_rate": 4.747874028753375e-07,
|
| 196 |
+
"logits/chosen": -0.240234375,
|
| 197 |
+
"logits/rejected": -0.2177734375,
|
| 198 |
+
"logps/chosen": -402.0,
|
| 199 |
+
"logps/rejected": -378.0,
|
| 200 |
+
"loss": 1.2237,
|
| 201 |
+
"rewards/accuracies": 0.6937500238418579,
|
| 202 |
+
"rewards/chosen": -0.2060546875,
|
| 203 |
+
"rewards/margins": 0.2412109375,
|
| 204 |
+
"rewards/rejected": -0.447265625,
|
| 205 |
+
"step": 110
|
| 206 |
+
},
|
| 207 |
+
{
|
| 208 |
+
"epoch": 0.2510460251046025,
|
| 209 |
+
"grad_norm": 14.2611531769557,
|
| 210 |
+
"learning_rate": 4.662012913161997e-07,
|
| 211 |
+
"logits/chosen": -0.244140625,
|
| 212 |
+
"logits/rejected": -0.265625,
|
| 213 |
+
"logps/chosen": -392.0,
|
| 214 |
+
"logps/rejected": -394.0,
|
| 215 |
+
"loss": 1.193,
|
| 216 |
+
"rewards/accuracies": 0.71875,
|
| 217 |
+
"rewards/chosen": -0.298828125,
|
| 218 |
+
"rewards/margins": 0.28125,
|
| 219 |
+
"rewards/rejected": -0.58203125,
|
| 220 |
+
"step": 120
|
| 221 |
+
},
|
| 222 |
+
{
|
| 223 |
+
"epoch": 0.2719665271966527,
|
| 224 |
+
"grad_norm": 12.798893628522647,
|
| 225 |
+
"learning_rate": 4.5646165232345103e-07,
|
| 226 |
+
"logits/chosen": -0.251953125,
|
| 227 |
+
"logits/rejected": -0.2490234375,
|
| 228 |
+
"logps/chosen": -358.0,
|
| 229 |
+
"logps/rejected": -396.0,
|
| 230 |
+
"loss": 1.18,
|
| 231 |
+
"rewards/accuracies": 0.731249988079071,
|
| 232 |
+
"rewards/chosen": -0.2158203125,
|
| 233 |
+
"rewards/margins": 0.359375,
|
| 234 |
+
"rewards/rejected": -0.578125,
|
| 235 |
+
"step": 130
|
| 236 |
+
},
|
| 237 |
+
{
|
| 238 |
+
"epoch": 0.2928870292887029,
|
| 239 |
+
"grad_norm": 19.239717506879277,
|
| 240 |
+
"learning_rate": 4.456204510851956e-07,
|
| 241 |
+
"logits/chosen": -0.29296875,
|
| 242 |
+
"logits/rejected": -0.263671875,
|
| 243 |
+
"logps/chosen": -400.0,
|
| 244 |
+
"logps/rejected": -428.0,
|
| 245 |
+
"loss": 1.1792,
|
| 246 |
+
"rewards/accuracies": 0.699999988079071,
|
| 247 |
+
"rewards/chosen": -0.2734375,
|
| 248 |
+
"rewards/margins": 0.37890625,
|
| 249 |
+
"rewards/rejected": -0.65234375,
|
| 250 |
+
"step": 140
|
| 251 |
+
},
|
| 252 |
+
{
|
| 253 |
+
"epoch": 0.3138075313807531,
|
| 254 |
+
"grad_norm": 18.538208596251096,
|
| 255 |
+
"learning_rate": 4.337355301007335e-07,
|
| 256 |
+
"logits/chosen": -0.287109375,
|
| 257 |
+
"logits/rejected": -0.259765625,
|
| 258 |
+
"logps/chosen": -380.0,
|
| 259 |
+
"logps/rejected": -414.0,
|
| 260 |
+
"loss": 1.1953,
|
| 261 |
+
"rewards/accuracies": 0.675000011920929,
|
| 262 |
+
"rewards/chosen": -0.3359375,
|
| 263 |
+
"rewards/margins": 0.26953125,
|
| 264 |
+
"rewards/rejected": -0.60546875,
|
| 265 |
+
"step": 150
|
| 266 |
+
},
|
| 267 |
+
{
|
| 268 |
+
"epoch": 0.33472803347280333,
|
| 269 |
+
"grad_norm": 13.726718415863484,
|
| 270 |
+
"learning_rate": 4.2087030056579986e-07,
|
| 271 |
+
"logits/chosen": -0.341796875,
|
| 272 |
+
"logits/rejected": -0.345703125,
|
| 273 |
+
"logps/chosen": -364.0,
|
| 274 |
+
"logps/rejected": -380.0,
|
| 275 |
+
"loss": 1.1963,
|
| 276 |
+
"rewards/accuracies": 0.6875,
|
| 277 |
+
"rewards/chosen": -0.2890625,
|
| 278 |
+
"rewards/margins": 0.373046875,
|
| 279 |
+
"rewards/rejected": -0.66015625,
|
| 280 |
+
"step": 160
|
| 281 |
+
},
|
| 282 |
+
{
|
| 283 |
+
"epoch": 0.35564853556485354,
|
| 284 |
+
"grad_norm": 15.071774554309869,
|
| 285 |
+
"learning_rate": 4.070934040463998e-07,
|
| 286 |
+
"logits/chosen": -0.2490234375,
|
| 287 |
+
"logits/rejected": -0.2265625,
|
| 288 |
+
"logps/chosen": -346.0,
|
| 289 |
+
"logps/rejected": -362.0,
|
| 290 |
+
"loss": 1.1709,
|
| 291 |
+
"rewards/accuracies": 0.6312500238418579,
|
| 292 |
+
"rewards/chosen": -0.34765625,
|
| 293 |
+
"rewards/margins": 0.31640625,
|
| 294 |
+
"rewards/rejected": -0.6640625,
|
| 295 |
+
"step": 170
|
| 296 |
+
},
|
| 297 |
+
{
|
| 298 |
+
"epoch": 0.37656903765690375,
|
| 299 |
+
"grad_norm": 14.777299895817363,
|
| 300 |
+
"learning_rate": 3.9247834624635404e-07,
|
| 301 |
+
"logits/chosen": -0.345703125,
|
| 302 |
+
"logits/rejected": -0.35546875,
|
| 303 |
+
"logps/chosen": -358.0,
|
| 304 |
+
"logps/rejected": -368.0,
|
| 305 |
+
"loss": 1.1282,
|
| 306 |
+
"rewards/accuracies": 0.7124999761581421,
|
| 307 |
+
"rewards/chosen": -0.359375,
|
| 308 |
+
"rewards/margins": 0.326171875,
|
| 309 |
+
"rewards/rejected": -0.68359375,
|
| 310 |
+
"step": 180
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"epoch": 0.39748953974895396,
|
| 314 |
+
"grad_norm": 17.917232829449304,
|
| 315 |
+
"learning_rate": 3.7710310482256523e-07,
|
| 316 |
+
"logits/chosen": -0.373046875,
|
| 317 |
+
"logits/rejected": -0.3359375,
|
| 318 |
+
"logps/chosen": -376.0,
|
| 319 |
+
"logps/rejected": -428.0,
|
| 320 |
+
"loss": 1.1659,
|
| 321 |
+
"rewards/accuracies": 0.7124999761581421,
|
| 322 |
+
"rewards/chosen": -0.30859375,
|
| 323 |
+
"rewards/margins": 0.482421875,
|
| 324 |
+
"rewards/rejected": -0.7890625,
|
| 325 |
+
"step": 190
|
| 326 |
+
},
|
| 327 |
+
{
|
| 328 |
+
"epoch": 0.41841004184100417,
|
| 329 |
+
"grad_norm": 26.129357821594134,
|
| 330 |
+
"learning_rate": 3.610497133404795e-07,
|
| 331 |
+
"logits/chosen": -0.328125,
|
| 332 |
+
"logits/rejected": -0.40234375,
|
| 333 |
+
"logps/chosen": -370.0,
|
| 334 |
+
"logps/rejected": -396.0,
|
| 335 |
+
"loss": 1.1624,
|
| 336 |
+
"rewards/accuracies": 0.762499988079071,
|
| 337 |
+
"rewards/chosen": -0.37890625,
|
| 338 |
+
"rewards/margins": 0.40625,
|
| 339 |
+
"rewards/rejected": -0.78515625,
|
| 340 |
+
"step": 200
|
| 341 |
+
},
|
| 342 |
+
{
|
| 343 |
+
"epoch": 0.41841004184100417,
|
| 344 |
+
"eval_logits/chosen": -0.400390625,
|
| 345 |
+
"eval_logits/rejected": -0.36328125,
|
| 346 |
+
"eval_logps/chosen": -364.0,
|
| 347 |
+
"eval_logps/rejected": -430.0,
|
| 348 |
+
"eval_loss": 0.5701406002044678,
|
| 349 |
+
"eval_rewards/accuracies": 0.7578125,
|
| 350 |
+
"eval_rewards/chosen": -0.271484375,
|
| 351 |
+
"eval_rewards/margins": 0.5546875,
|
| 352 |
+
"eval_rewards/rejected": -0.828125,
|
| 353 |
+
"eval_runtime": 44.3219,
|
| 354 |
+
"eval_samples_per_second": 45.124,
|
| 355 |
+
"eval_steps_per_second": 0.722,
|
| 356 |
+
"step": 200
|
| 357 |
+
},
|
| 358 |
+
{
|
| 359 |
+
"epoch": 0.4393305439330544,
|
| 360 |
+
"grad_norm": 20.166611438012268,
|
| 361 |
+
"learning_rate": 3.4440382358952115e-07,
|
| 362 |
+
"logits/chosen": -0.44140625,
|
| 363 |
+
"logits/rejected": -0.427734375,
|
| 364 |
+
"logps/chosen": -388.0,
|
| 365 |
+
"logps/rejected": -406.0,
|
| 366 |
+
"loss": 1.166,
|
| 367 |
+
"rewards/accuracies": 0.7437499761581421,
|
| 368 |
+
"rewards/chosen": -0.333984375,
|
| 369 |
+
"rewards/margins": 0.462890625,
|
| 370 |
+
"rewards/rejected": -0.796875,
|
| 371 |
+
"step": 210
|
| 372 |
+
},
|
| 373 |
+
{
|
| 374 |
+
"epoch": 0.4602510460251046,
|
| 375 |
+
"grad_norm": 20.143454486690988,
|
| 376 |
+
"learning_rate": 3.272542485937368e-07,
|
| 377 |
+
"logits/chosen": -0.43359375,
|
| 378 |
+
"logits/rejected": -0.4921875,
|
| 379 |
+
"logps/chosen": -366.0,
|
| 380 |
+
"logps/rejected": -408.0,
|
| 381 |
+
"loss": 1.1348,
|
| 382 |
+
"rewards/accuracies": 0.7250000238418579,
|
| 383 |
+
"rewards/chosen": -0.37890625,
|
| 384 |
+
"rewards/margins": 0.50390625,
|
| 385 |
+
"rewards/rejected": -0.8828125,
|
| 386 |
+
"step": 220
|
| 387 |
+
},
|
| 388 |
+
{
|
| 389 |
+
"epoch": 0.4811715481171548,
|
| 390 |
+
"grad_norm": 21.218520645945883,
|
| 391 |
+
"learning_rate": 3.096924887558854e-07,
|
| 392 |
+
"logits/chosen": -0.3515625,
|
| 393 |
+
"logits/rejected": -0.404296875,
|
| 394 |
+
"logps/chosen": -352.0,
|
| 395 |
+
"logps/rejected": -408.0,
|
| 396 |
+
"loss": 1.1448,
|
| 397 |
+
"rewards/accuracies": 0.7437499761581421,
|
| 398 |
+
"rewards/chosen": -0.443359375,
|
| 399 |
+
"rewards/margins": 0.515625,
|
| 400 |
+
"rewards/rejected": -0.9609375,
|
| 401 |
+
"step": 230
|
| 402 |
+
},
|
| 403 |
+
{
|
| 404 |
+
"epoch": 0.502092050209205,
|
| 405 |
+
"grad_norm": 16.293281058848866,
|
| 406 |
+
"learning_rate": 2.9181224366319943e-07,
|
| 407 |
+
"logits/chosen": -0.3828125,
|
| 408 |
+
"logits/rejected": -0.333984375,
|
| 409 |
+
"logps/chosen": -366.0,
|
| 410 |
+
"logps/rejected": -414.0,
|
| 411 |
+
"loss": 1.1016,
|
| 412 |
+
"rewards/accuracies": 0.78125,
|
| 413 |
+
"rewards/chosen": -0.375,
|
| 414 |
+
"rewards/margins": 0.51171875,
|
| 415 |
+
"rewards/rejected": -0.88671875,
|
| 416 |
+
"step": 240
|
| 417 |
+
},
|
| 418 |
+
{
|
| 419 |
+
"epoch": 0.5230125523012552,
|
| 420 |
+
"grad_norm": 14.934923417125477,
|
| 421 |
+
"learning_rate": 2.7370891215954565e-07,
|
| 422 |
+
"logits/chosen": -0.470703125,
|
| 423 |
+
"logits/rejected": -0.40234375,
|
| 424 |
+
"logps/chosen": -386.0,
|
| 425 |
+
"logps/rejected": -422.0,
|
| 426 |
+
"loss": 1.1298,
|
| 427 |
+
"rewards/accuracies": 0.731249988079071,
|
| 428 |
+
"rewards/chosen": -0.310546875,
|
| 429 |
+
"rewards/margins": 0.58203125,
|
| 430 |
+
"rewards/rejected": -0.890625,
|
| 431 |
+
"step": 250
|
| 432 |
+
},
|
| 433 |
+
{
|
| 434 |
+
"epoch": 0.5439330543933054,
|
| 435 |
+
"grad_norm": 17.722282481417537,
|
| 436 |
+
"learning_rate": 2.55479083351317e-07,
|
| 437 |
+
"logits/chosen": -0.5,
|
| 438 |
+
"logits/rejected": -0.48046875,
|
| 439 |
+
"logps/chosen": -392.0,
|
| 440 |
+
"logps/rejected": -418.0,
|
| 441 |
+
"loss": 1.1069,
|
| 442 |
+
"rewards/accuracies": 0.7562500238418579,
|
| 443 |
+
"rewards/chosen": -0.376953125,
|
| 444 |
+
"rewards/margins": 0.58203125,
|
| 445 |
+
"rewards/rejected": -0.9609375,
|
| 446 |
+
"step": 260
|
| 447 |
+
},
|
| 448 |
+
{
|
| 449 |
+
"epoch": 0.5648535564853556,
|
| 450 |
+
"grad_norm": 15.709708500953589,
|
| 451 |
+
"learning_rate": 2.3722002126275822e-07,
|
| 452 |
+
"logits/chosen": -0.404296875,
|
| 453 |
+
"logits/rejected": -0.40625,
|
| 454 |
+
"logps/chosen": -394.0,
|
| 455 |
+
"logps/rejected": -414.0,
|
| 456 |
+
"loss": 1.1279,
|
| 457 |
+
"rewards/accuracies": 0.6625000238418579,
|
| 458 |
+
"rewards/chosen": -0.423828125,
|
| 459 |
+
"rewards/margins": 0.427734375,
|
| 460 |
+
"rewards/rejected": -0.8515625,
|
| 461 |
+
"step": 270
|
| 462 |
+
},
|
| 463 |
+
{
|
| 464 |
+
"epoch": 0.5857740585774058,
|
| 465 |
+
"grad_norm": 22.495247525440202,
|
| 466 |
+
"learning_rate": 2.19029145890313e-07,
|
| 467 |
+
"logits/chosen": -0.36328125,
|
| 468 |
+
"logits/rejected": -0.37109375,
|
| 469 |
+
"logps/chosen": -370.0,
|
| 470 |
+
"logps/rejected": -402.0,
|
| 471 |
+
"loss": 1.1459,
|
| 472 |
+
"rewards/accuracies": 0.65625,
|
| 473 |
+
"rewards/chosen": -0.369140625,
|
| 474 |
+
"rewards/margins": 0.51171875,
|
| 475 |
+
"rewards/rejected": -0.8828125,
|
| 476 |
+
"step": 280
|
| 477 |
+
},
|
| 478 |
+
{
|
| 479 |
+
"epoch": 0.606694560669456,
|
| 480 |
+
"grad_norm": 17.685639076240154,
|
| 481 |
+
"learning_rate": 2.0100351342479216e-07,
|
| 482 |
+
"logits/chosen": -0.5,
|
| 483 |
+
"logits/rejected": -0.423828125,
|
| 484 |
+
"logps/chosen": -352.0,
|
| 485 |
+
"logps/rejected": -400.0,
|
| 486 |
+
"loss": 1.1235,
|
| 487 |
+
"rewards/accuracies": 0.675000011920929,
|
| 488 |
+
"rewards/chosen": -0.412109375,
|
| 489 |
+
"rewards/margins": 0.474609375,
|
| 490 |
+
"rewards/rejected": -0.88671875,
|
| 491 |
+
"step": 290
|
| 492 |
+
},
|
| 493 |
+
{
|
| 494 |
+
"epoch": 0.6276150627615062,
|
| 495 |
+
"grad_norm": 15.591635531747432,
|
| 496 |
+
"learning_rate": 1.8323929841460178e-07,
|
| 497 |
+
"logits/chosen": -0.466796875,
|
| 498 |
+
"logits/rejected": -0.470703125,
|
| 499 |
+
"logps/chosen": -418.0,
|
| 500 |
+
"logps/rejected": -432.0,
|
| 501 |
+
"loss": 1.0945,
|
| 502 |
+
"rewards/accuracies": 0.7124999761581421,
|
| 503 |
+
"rewards/chosen": -0.5,
|
| 504 |
+
"rewards/margins": 0.47265625,
|
| 505 |
+
"rewards/rejected": -0.97265625,
|
| 506 |
+
"step": 300
|
| 507 |
+
},
|
| 508 |
+
{
|
| 509 |
+
"epoch": 0.6276150627615062,
|
| 510 |
+
"eval_logits/chosen": -0.453125,
|
| 511 |
+
"eval_logits/rejected": -0.427734375,
|
| 512 |
+
"eval_logps/chosen": -368.0,
|
| 513 |
+
"eval_logps/rejected": -444.0,
|
| 514 |
+
"eval_loss": 0.552929699420929,
|
| 515 |
+
"eval_rewards/accuracies": 0.765625,
|
| 516 |
+
"eval_rewards/chosen": -0.318359375,
|
| 517 |
+
"eval_rewards/margins": 0.65234375,
|
| 518 |
+
"eval_rewards/rejected": -0.96875,
|
| 519 |
+
"eval_runtime": 44.2434,
|
| 520 |
+
"eval_samples_per_second": 45.204,
|
| 521 |
+
"eval_steps_per_second": 0.723,
|
| 522 |
+
"step": 300
|
| 523 |
+
},
|
| 524 |
+
{
|
| 525 |
+
"epoch": 0.6485355648535565,
|
| 526 |
+
"grad_norm": 19.76547371178574,
|
| 527 |
+
"learning_rate": 1.6583128063291573e-07,
|
| 528 |
+
"logits/chosen": -0.341796875,
|
| 529 |
+
"logits/rejected": -0.421875,
|
| 530 |
+
"logps/chosen": -408.0,
|
| 531 |
+
"logps/rejected": -428.0,
|
| 532 |
+
"loss": 1.0912,
|
| 533 |
+
"rewards/accuracies": 0.71875,
|
| 534 |
+
"rewards/chosen": -0.47265625,
|
| 535 |
+
"rewards/margins": 0.55078125,
|
| 536 |
+
"rewards/rejected": -1.0234375,
|
| 537 |
+
"step": 310
|
| 538 |
+
},
|
| 539 |
+
{
|
| 540 |
+
"epoch": 0.6694560669456067,
|
| 541 |
+
"grad_norm": 21.414058995151994,
|
| 542 |
+
"learning_rate": 1.488723393865766e-07,
|
| 543 |
+
"logits/chosen": -0.482421875,
|
| 544 |
+
"logits/rejected": -0.4609375,
|
| 545 |
+
"logps/chosen": -400.0,
|
| 546 |
+
"logps/rejected": -422.0,
|
| 547 |
+
"loss": 1.0604,
|
| 548 |
+
"rewards/accuracies": 0.75,
|
| 549 |
+
"rewards/chosen": -0.416015625,
|
| 550 |
+
"rewards/margins": 0.62890625,
|
| 551 |
+
"rewards/rejected": -1.046875,
|
| 552 |
+
"step": 320
|
| 553 |
+
},
|
| 554 |
+
{
|
| 555 |
+
"epoch": 0.6903765690376569,
|
| 556 |
+
"grad_norm": 19.414772003817266,
|
| 557 |
+
"learning_rate": 1.3245295796480788e-07,
|
| 558 |
+
"logits/chosen": -0.51171875,
|
| 559 |
+
"logits/rejected": -0.498046875,
|
| 560 |
+
"logps/chosen": -374.0,
|
| 561 |
+
"logps/rejected": -430.0,
|
| 562 |
+
"loss": 1.1006,
|
| 563 |
+
"rewards/accuracies": 0.706250011920929,
|
| 564 |
+
"rewards/chosen": -0.46484375,
|
| 565 |
+
"rewards/margins": 0.451171875,
|
| 566 |
+
"rewards/rejected": -0.9140625,
|
| 567 |
+
"step": 330
|
| 568 |
+
},
|
| 569 |
+
{
|
| 570 |
+
"epoch": 0.7112970711297071,
|
| 571 |
+
"grad_norm": 21.720939162926207,
|
| 572 |
+
"learning_rate": 1.1666074087171627e-07,
|
| 573 |
+
"logits/chosen": -0.52734375,
|
| 574 |
+
"logits/rejected": -0.5078125,
|
| 575 |
+
"logps/chosen": -408.0,
|
| 576 |
+
"logps/rejected": -472.0,
|
| 577 |
+
"loss": 1.0815,
|
| 578 |
+
"rewards/accuracies": 0.7875000238418579,
|
| 579 |
+
"rewards/chosen": -0.474609375,
|
| 580 |
+
"rewards/margins": 0.70703125,
|
| 581 |
+
"rewards/rejected": -1.1796875,
|
| 582 |
+
"step": 340
|
| 583 |
+
},
|
| 584 |
+
{
|
| 585 |
+
"epoch": 0.7322175732217573,
|
| 586 |
+
"grad_norm": 20.08927723881005,
|
| 587 |
+
"learning_rate": 1.0157994641835734e-07,
|
| 588 |
+
"logits/chosen": -0.45703125,
|
| 589 |
+
"logits/rejected": -0.474609375,
|
| 590 |
+
"logps/chosen": -376.0,
|
| 591 |
+
"logps/rejected": -416.0,
|
| 592 |
+
"loss": 1.0618,
|
| 593 |
+
"rewards/accuracies": 0.7124999761581421,
|
| 594 |
+
"rewards/chosen": -0.462890625,
|
| 595 |
+
"rewards/margins": 0.62109375,
|
| 596 |
+
"rewards/rejected": -1.0859375,
|
| 597 |
+
"step": 350
|
| 598 |
+
},
|
| 599 |
+
{
|
| 600 |
+
"epoch": 0.7531380753138075,
|
| 601 |
+
"grad_norm": 19.29218318311807,
|
| 602 |
+
"learning_rate": 8.729103716819111e-08,
|
| 603 |
+
"logits/chosen": -0.515625,
|
| 604 |
+
"logits/rejected": -0.4453125,
|
| 605 |
+
"logps/chosen": -424.0,
|
| 606 |
+
"logps/rejected": -448.0,
|
| 607 |
+
"loss": 1.1298,
|
| 608 |
+
"rewards/accuracies": 0.699999988079071,
|
| 609 |
+
"rewards/chosen": -0.53515625,
|
| 610 |
+
"rewards/margins": 0.60546875,
|
| 611 |
+
"rewards/rejected": -1.140625,
|
| 612 |
+
"step": 360
|
| 613 |
+
},
|
| 614 |
+
{
|
| 615 |
+
"epoch": 0.7740585774058577,
|
| 616 |
+
"grad_norm": 18.524290784854816,
|
| 617 |
+
"learning_rate": 7.387025063449081e-08,
|
| 618 |
+
"logits/chosen": -0.48046875,
|
| 619 |
+
"logits/rejected": -0.50390625,
|
| 620 |
+
"logps/chosen": -390.0,
|
| 621 |
+
"logps/rejected": -406.0,
|
| 622 |
+
"loss": 1.0897,
|
| 623 |
+
"rewards/accuracies": 0.731249988079071,
|
| 624 |
+
"rewards/chosen": -0.5390625,
|
| 625 |
+
"rewards/margins": 0.53125,
|
| 626 |
+
"rewards/rejected": -1.0703125,
|
| 627 |
+
"step": 370
|
| 628 |
+
},
|
| 629 |
+
{
|
| 630 |
+
"epoch": 0.7949790794979079,
|
| 631 |
+
"grad_norm": 17.245854433432992,
|
| 632 |
+
"learning_rate": 6.138919252022435e-08,
|
| 633 |
+
"logits/chosen": -0.392578125,
|
| 634 |
+
"logits/rejected": -0.404296875,
|
| 635 |
+
"logps/chosen": -360.0,
|
| 636 |
+
"logps/rejected": -456.0,
|
| 637 |
+
"loss": 1.0967,
|
| 638 |
+
"rewards/accuracies": 0.737500011920929,
|
| 639 |
+
"rewards/chosen": -0.5703125,
|
| 640 |
+
"rewards/margins": 0.7109375,
|
| 641 |
+
"rewards/rejected": -1.28125,
|
| 642 |
+
"step": 380
|
| 643 |
+
},
|
| 644 |
+
{
|
| 645 |
+
"epoch": 0.8158995815899581,
|
| 646 |
+
"grad_norm": 25.69229657215138,
|
| 647 |
+
"learning_rate": 4.991445467064689e-08,
|
| 648 |
+
"logits/chosen": -0.51171875,
|
| 649 |
+
"logits/rejected": -0.4609375,
|
| 650 |
+
"logps/chosen": -424.0,
|
| 651 |
+
"logps/rejected": -474.0,
|
| 652 |
+
"loss": 1.0782,
|
| 653 |
+
"rewards/accuracies": 0.7250000238418579,
|
| 654 |
+
"rewards/chosen": -0.53515625,
|
| 655 |
+
"rewards/margins": 0.609375,
|
| 656 |
+
"rewards/rejected": -1.1484375,
|
| 657 |
+
"step": 390
|
| 658 |
+
},
|
| 659 |
+
{
|
| 660 |
+
"epoch": 0.8368200836820083,
|
| 661 |
+
"grad_norm": 34.70688710928117,
|
| 662 |
+
"learning_rate": 3.9507259776993954e-08,
|
| 663 |
+
"logits/chosen": -0.51953125,
|
| 664 |
+
"logits/rejected": -0.466796875,
|
| 665 |
+
"logps/chosen": -390.0,
|
| 666 |
+
"logps/rejected": -456.0,
|
| 667 |
+
"loss": 1.0971,
|
| 668 |
+
"rewards/accuracies": 0.737500011920929,
|
| 669 |
+
"rewards/chosen": -0.5234375,
|
| 670 |
+
"rewards/margins": 0.6328125,
|
| 671 |
+
"rewards/rejected": -1.15625,
|
| 672 |
+
"step": 400
|
| 673 |
+
},
|
| 674 |
+
{
|
| 675 |
+
"epoch": 0.8368200836820083,
|
| 676 |
+
"eval_logits/chosen": -0.4921875,
|
| 677 |
+
"eval_logits/rejected": -0.47265625,
|
| 678 |
+
"eval_logps/chosen": -376.0,
|
| 679 |
+
"eval_logps/rejected": -460.0,
|
| 680 |
+
"eval_loss": 0.5477031469345093,
|
| 681 |
+
"eval_rewards/accuracies": 0.75390625,
|
| 682 |
+
"eval_rewards/chosen": -0.408203125,
|
| 683 |
+
"eval_rewards/margins": 0.7265625,
|
| 684 |
+
"eval_rewards/rejected": -1.1328125,
|
| 685 |
+
"eval_runtime": 44.1617,
|
| 686 |
+
"eval_samples_per_second": 45.288,
|
| 687 |
+
"eval_steps_per_second": 0.725,
|
| 688 |
+
"step": 400
|
| 689 |
+
},
|
| 690 |
+
{
|
| 691 |
+
"epoch": 0.8577405857740585,
|
| 692 |
+
"grad_norm": 19.643118113308155,
|
| 693 |
+
"learning_rate": 3.022313472693447e-08,
|
| 694 |
+
"logits/chosen": -0.515625,
|
| 695 |
+
"logits/rejected": -0.515625,
|
| 696 |
+
"logps/chosen": -398.0,
|
| 697 |
+
"logps/rejected": -442.0,
|
| 698 |
+
"loss": 1.101,
|
| 699 |
+
"rewards/accuracies": 0.7250000238418579,
|
| 700 |
+
"rewards/chosen": -0.47265625,
|
| 701 |
+
"rewards/margins": 0.578125,
|
| 702 |
+
"rewards/rejected": -1.0546875,
|
| 703 |
+
"step": 410
|
| 704 |
+
},
|
| 705 |
+
{
|
| 706 |
+
"epoch": 0.8786610878661087,
|
| 707 |
+
"grad_norm": 19.28458618388238,
|
| 708 |
+
"learning_rate": 2.2111614344599684e-08,
|
| 709 |
+
"logits/chosen": -0.4453125,
|
| 710 |
+
"logits/rejected": -0.498046875,
|
| 711 |
+
"logps/chosen": -438.0,
|
| 712 |
+
"logps/rejected": -480.0,
|
| 713 |
+
"loss": 1.0719,
|
| 714 |
+
"rewards/accuracies": 0.6937500238418579,
|
| 715 |
+
"rewards/chosen": -0.52734375,
|
| 716 |
+
"rewards/margins": 0.6015625,
|
| 717 |
+
"rewards/rejected": -1.1328125,
|
| 718 |
+
"step": 420
|
| 719 |
+
},
|
| 720 |
+
{
|
| 721 |
+
"epoch": 0.899581589958159,
|
| 722 |
+
"grad_norm": 16.558119542768463,
|
| 723 |
+
"learning_rate": 1.521597710086439e-08,
|
| 724 |
+
"logits/chosen": -0.4453125,
|
| 725 |
+
"logits/rejected": -0.408203125,
|
| 726 |
+
"logps/chosen": -416.0,
|
| 727 |
+
"logps/rejected": -444.0,
|
| 728 |
+
"loss": 1.0792,
|
| 729 |
+
"rewards/accuracies": 0.71875,
|
| 730 |
+
"rewards/chosen": -0.5859375,
|
| 731 |
+
"rewards/margins": 0.58203125,
|
| 732 |
+
"rewards/rejected": -1.1640625,
|
| 733 |
+
"step": 430
|
| 734 |
+
},
|
| 735 |
+
{
|
| 736 |
+
"epoch": 0.9205020920502092,
|
| 737 |
+
"grad_norm": 19.88416183093778,
|
| 738 |
+
"learning_rate": 9.57301420397924e-09,
|
| 739 |
+
"logits/chosen": -0.466796875,
|
| 740 |
+
"logits/rejected": -0.462890625,
|
| 741 |
+
"logps/chosen": -404.0,
|
| 742 |
+
"logps/rejected": -440.0,
|
| 743 |
+
"loss": 1.097,
|
| 744 |
+
"rewards/accuracies": 0.668749988079071,
|
| 745 |
+
"rewards/chosen": -0.5,
|
| 746 |
+
"rewards/margins": 0.515625,
|
| 747 |
+
"rewards/rejected": -1.015625,
|
| 748 |
+
"step": 440
|
| 749 |
+
},
|
| 750 |
+
{
|
| 751 |
+
"epoch": 0.9414225941422594,
|
| 752 |
+
"grad_norm": 16.17448723735275,
|
| 753 |
+
"learning_rate": 5.212833302556258e-09,
|
| 754 |
+
"logits/chosen": -0.546875,
|
| 755 |
+
"logits/rejected": -0.5625,
|
| 756 |
+
"logps/chosen": -400.0,
|
| 757 |
+
"logps/rejected": -486.0,
|
| 758 |
+
"loss": 1.0946,
|
| 759 |
+
"rewards/accuracies": 0.7250000238418579,
|
| 760 |
+
"rewards/chosen": -0.52734375,
|
| 761 |
+
"rewards/margins": 0.5625,
|
| 762 |
+
"rewards/rejected": -1.0859375,
|
| 763 |
+
"step": 450
|
| 764 |
+
},
|
| 765 |
+
{
|
| 766 |
+
"epoch": 0.9623430962343096,
|
| 767 |
+
"grad_norm": 25.24493230543812,
|
| 768 |
+
"learning_rate": 2.158697848236607e-09,
|
| 769 |
+
"logits/chosen": -0.46875,
|
| 770 |
+
"logits/rejected": -0.5,
|
| 771 |
+
"logps/chosen": -386.0,
|
| 772 |
+
"logps/rejected": -414.0,
|
| 773 |
+
"loss": 1.0964,
|
| 774 |
+
"rewards/accuracies": 0.71875,
|
| 775 |
+
"rewards/chosen": -0.447265625,
|
| 776 |
+
"rewards/margins": 0.5546875,
|
| 777 |
+
"rewards/rejected": -1.0,
|
| 778 |
+
"step": 460
|
| 779 |
+
},
|
| 780 |
+
{
|
| 781 |
+
"epoch": 0.9832635983263598,
|
| 782 |
+
"grad_norm": 17.43448970223912,
|
| 783 |
+
"learning_rate": 4.269029751107489e-10,
|
| 784 |
+
"logits/chosen": -0.439453125,
|
| 785 |
+
"logits/rejected": -0.49609375,
|
| 786 |
+
"logps/chosen": -396.0,
|
| 787 |
+
"logps/rejected": -456.0,
|
| 788 |
+
"loss": 1.0763,
|
| 789 |
+
"rewards/accuracies": 0.7250000238418579,
|
| 790 |
+
"rewards/chosen": -0.50390625,
|
| 791 |
+
"rewards/margins": 0.515625,
|
| 792 |
+
"rewards/rejected": -1.0234375,
|
| 793 |
+
"step": 470
|
| 794 |
+
},
|
| 795 |
+
{
|
| 796 |
+
"epoch": 1.0,
|
| 797 |
+
"step": 478,
|
| 798 |
+
"total_flos": 0.0,
|
| 799 |
+
"train_loss": 1.1650768583289748,
|
| 800 |
+
"train_runtime": 3769.2146,
|
| 801 |
+
"train_samples_per_second": 16.219,
|
| 802 |
+
"train_steps_per_second": 0.127
|
| 803 |
+
}
|
| 804 |
+
],
|
| 805 |
+
"logging_steps": 10,
|
| 806 |
+
"max_steps": 478,
|
| 807 |
+
"num_input_tokens_seen": 0,
|
| 808 |
+
"num_train_epochs": 1,
|
| 809 |
+
"save_steps": 100,
|
| 810 |
+
"stateful_callbacks": {
|
| 811 |
+
"TrainerControl": {
|
| 812 |
+
"args": {
|
| 813 |
+
"should_epoch_stop": false,
|
| 814 |
+
"should_evaluate": false,
|
| 815 |
+
"should_log": false,
|
| 816 |
+
"should_save": true,
|
| 817 |
+
"should_training_stop": true
|
| 818 |
+
},
|
| 819 |
+
"attributes": {}
|
| 820 |
+
}
|
| 821 |
+
},
|
| 822 |
+
"total_flos": 0.0,
|
| 823 |
+
"train_batch_size": 8,
|
| 824 |
+
"trial_name": null,
|
| 825 |
+
"trial_params": null
|
| 826 |
+
}
|