|
--- |
|
base_model: Qwen/Qwen2-0.5B-Instruct |
|
datasets: trl-lib/ultrafeedback-prompt |
|
library_name: transformers |
|
model_name: xpo-qwen2 |
|
tags: |
|
- trl |
|
- generated_from_trainer |
|
- xpo |
|
licence: license |
|
--- |
|
|
|
# Model Card for xpo-qwen2 |
|
|
|
This model is a fine-tuned version of [Qwen/Qwen2-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2-0.5B-Instruct) on the [trl-lib/ultrafeedback-prompt](https://huggingface.co/datasets/trl-lib/ultrafeedback-prompt) 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="qgallouedec/xpo-qwen2", device="cuda") |
|
output = generator([{"role": "user", "content": question}], max_new_tokens=500)[0] |
|
print(output["generated_text"][1]["content"]) |
|
``` |
|
|
|
## 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/huggingface/huggingface/runs/bg6y6mom) |
|
|
|
This model was trained with XPO, a method introduced in [Exploratory Preference Optimization: Harnessing Implicit Q*-Approximation for Sample-Efficient RLHF](https://huggingface.co/papers/2405.21046). |
|
|
|
### Framework versions |
|
|
|
- TRL: 0.12.0.dev0 |
|
- Transformers: 4.45.0.dev0 |
|
- Pytorch: 2.4.1 |
|
- Datasets: 3.0.0 |
|
- Tokenizers: 0.19.1 |
|
|
|
## Citations |
|
|
|
Cite XPO as: |
|
|
|
```bibtex |
|
@article{jung2024binary, |
|
title = {{Binary Classifier Optimization for Large Language Model Alignment}}, |
|
author = {Seungjae Jung and Gunsoo Han and Daniel Wontae Nam and Kyoung{-}Woon On}, |
|
year = 2024, |
|
eprint = {arXiv:2404.04656} |
|
} |
|
``` |
|
|
|
Cite TRL as: |
|
|
|
```bibtex |
|
@misc{vonwerra2022trl, |
|
title = {{TRL: Transformer Reinforcement Learning}}, |
|
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}, |
|
year = 2020, |
|
journal = {GitHub repository}, |
|
publisher = {GitHub}, |
|
howpublished = {\url{https://github.com/huggingface/trl}} |
|
} |
|
``` |