59f90b518b
Model: code-critic-model/Qwen3-4B-SFT-DPO-4B-1409i-beta0.15-sft0.25-lr1e-6-bs32-ep3 Source: Original Platform
75 lines
3.2 KiB
Markdown
75 lines
3.2 KiB
Markdown
---
|
|
base_model: code-critic-model/qwen3-4b-sft-prm
|
|
datasets: code-critic-model/4B-1409i
|
|
library_name: transformers
|
|
model_name: Qwen3-4B-SFT-DPO-4B-1409i-beta0.15-sft0.25-lr1e-6-bs32-ep3
|
|
tags:
|
|
- generated_from_trainer
|
|
- code-critic
|
|
- trl
|
|
- process-reward-model
|
|
- dpo
|
|
licence: license
|
|
---
|
|
|
|
> **Not used in the paper.** This is a development checkpoint kept so that existing links keep working. The critics reported in [Steer, Don't Solve](https://arxiv.org/abs/2606.21811) are listed on the [organization page](https://huggingface.co/code-critic-model); the released DPO critic is [Qwen3-4B-Critic-SFT-DPO](https://huggingface.co/code-critic-model/Qwen3-4B-Critic-SFT-DPO).
|
|
|
|
|
|
# Model Card for Qwen3-4B-SFT-DPO-4B-1409i-beta0.15-sft0.25-lr1e-6-bs32-ep3
|
|
|
|
This model is a fine-tuned version of [code-critic-model/qwen3-4b-sft-prm](https://huggingface.co/code-critic-model/qwen3-4b-sft-prm) on the [code-critic-model/4B-1409i](https://huggingface.co/datasets/code-critic-model/4B-1409i) dataset.
|
|
It has been trained using [TRL](https://github.com/huggingface/trl).
|
|
|
|
## Quick start
|
|
|
|
```python
|
|
from transformers import pipeline
|
|
|
|
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?"
|
|
generator = pipeline("text-generation", model="code-critic-model/Qwen3-4B-SFT-DPO-4B-1409i-beta0.15-sft0.25-lr1e-6-bs32-ep3", device="cuda")
|
|
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
|
|
print(output["generated_text"])
|
|
```
|
|
|
|
## Training procedure
|
|
|
|
[<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/code-critic-model/code-critic-dpo/runs/6bmarwrl)
|
|
|
|
|
|
|
|
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).
|
|
|
|
### Framework versions
|
|
|
|
- TRL: 1.11.0.dev0
|
|
- Transformers: 5.14.1
|
|
- Pytorch: 2.10.0+cu128
|
|
- Datasets: 5.0.1
|
|
- Tokenizers: 0.22.2
|
|
|
|
## Citations
|
|
|
|
Cite DPO as:
|
|
|
|
```bibtex
|
|
@inproceedings{rafailov2023direct,
|
|
title = {{Direct Preference Optimization: Your Language Model is Secretly a Reward Model}},
|
|
author = {Rafael Rafailov and Archit Sharma and Eric Mitchell and Christopher D. Manning and Stefano Ermon and Chelsea Finn},
|
|
year = 2023,
|
|
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},
|
|
url = {http://papers.nips.cc/paper_files/paper/2023/hash/a85b405ed65c6477a4fe8302b5e06ce7-Abstract-Conference.html},
|
|
editor = {Alice Oh and Tristan Naumann and Amir Globerson and Kate Saenko and Moritz Hardt and Sergey Levine},
|
|
}
|
|
```
|
|
|
|
Cite TRL as:
|
|
|
|
```bibtex
|
|
@software{vonwerra2020trl,
|
|
title = {{TRL: Transformers Reinforcement Learning}},
|
|
author = {von Werra, Leandro and Belkada, Younes and Tunstall, Lewis and Beeching, Edward and Thrush, Tristan and Lambert, Nathan and Huang, Shengyi and Rasul, Kashif and Gallouédec, Quentin},
|
|
license = {Apache-2.0},
|
|
url = {https://github.com/huggingface/trl},
|
|
year = {2020}
|
|
}
|
|
``` |