Available sizes

Branch	Bits	Description
8_0	8.0	Maximum quality that ExLlamaV2 can produce, near unquantized performance.
6_5	6.5	Very similar to 8.0, good tradeoff of size vs performance, recommended.
5_0	5.0	Slightly lower quality vs 6.5, but usable
4_25	4.25	GPTQ equivalent bits per weight, slightly higher quality.
3_5	3.5	Lower quality, only use if you have to.

Download instructions

With git:

git clone --single-branch --branch 6_5 https://huggingface.co/trl-lib_-_Qwen2-0.5B-XPO-exl2 Qwen2-0.5B-XPO-6_5

With huggingface hub:

pip3 install huggingface-hub

To download a specific branch, use the --revision parameter. For example, to download the 6.5 bpw branch: Linux:

huggingface-cli download trl-lib_-_Qwen2-0.5B-XPO-exl2 --revision 6_5 --local-dir Qwen2-0.5B-XPO-6_5 --local-dir-use-symlinks False

Windows (which apparently doesn't like _ in folders sometimes?):

huggingface-cli download trl-lib_-_Qwen2-0.5B-XPO-exl2 --revision 6_5 --local-dir Qwen2-0.5B-XPO-6.5 --local-dir-use-symlinks False

Original model description:

base_model: Qwen/Qwen2-0.5B-Instruct library_name: transformers model_name: Qwen2-0.5B-XPO tags: - generated_from_trainer - trl - xpo licence: license

Model Card for Qwen2-0.5B-XPO

This model is a fine-tuned version of Qwen/Qwen2-0.5B-Instruct. It has been trained using TRL.

Quick start

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/Qwen2-0.5B-XPO", device="cuda")
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
print(output["generated_text"])

Training procedure

This model was trained with XPO, a method introduced in Exploratory Preference Optimization: Harnessing Implicit Q*-Approximation for Sample-Efficient RLHF.

Framework versions

TRL: 0.12.0.dev0
Transformers: 4.46.0.dev0
Pytorch: 2.4.1
Datasets: 3.0.1
Tokenizers: 0.20.0

Citations

Cite XPO as:

@article{jung2024binary,
    title        = {{Exploratory Preference Optimization: Harnessing Implicit Q*-Approximation for Sample-Efficient RLHF}},
    author       = {Tengyang Xie and Dylan J. Foster and Akshay Krishnamurthy and Corby Rosset and Ahmed Awadallah and Alexander Rakhlin},
    year         = 2024,
    eprint       = {arXiv:2405.21046}
}

Cite TRL as:

@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}}
}