Model Card for pythia-1b-drpo-lora-tldr

This model is a fine-tuned version of cleanrl/EleutherAI_pythia-1b-deduped__sft__tldr. 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="Eehan/pythia-1b-drpo-lora-tldr", 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 DRPO, a method introduced in Doubly Robust Alignment for Large Language Models.

Framework versions

• TRL: 0.19.1
• Transformers: 4.54.0
• Pytorch: 2.7.1
• Datasets: 4.0.0
• Tokenizers: 0.21.2

Citations

Cite DRPO as:

@article{xu2024doubly,
    title        = {{Doubly Robust Alignment for Large Language Models}},
    author       = {Xu, Erhan and Ye, Kai and Zhou, Hongyi and Zhu, Luhan and Quinzan, Francesco and Shi, Chengchun},
    year         = 2025,
    journal      = {arXiv preprint arXiv:2506.01183}
}

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 Gallou{\'e}dec},
    year         = 2020,
    journal      = {GitHub repository},
    publisher    = {GitHub},
    howpublished = {\url{https://github.com/huggingface/trl}}
}