qwen3-4b-dw-lr-hf-dpo/README.md

base_model, datasets, library_name, model_name, tags, licence

base_model	datasets	library_name	model_name	tags	licence
cemig-temp/qwen3-4b-dw-lr	CEIA-RL/Nemotron-SFT-Safety-pt-BR-Cleaned	transformers	qwen3-4b-dw-lr-hf-dpo	generated_from_trainer trl online-dpo	license

Model Card for qwen3-4b-dw-lr-hf-dpo

This model is a fine-tuned version of cemig-temp/qwen3-4b-dw-lr on the CEIA-RL/Nemotron-SFT-Safety-pt-BR-Cleaned dataset. 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="CEIA-RL/qwen3-4b-dw-lr-hf-dpo", 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 Online DPO, a method introduced in Direct Language Model Alignment from Online AI Feedback.

Framework versions

• TRL: 0.29.0
• Transformers: 4.57.6
• Pytorch: 2.10.0
• Datasets: 4.7.0
• Tokenizers: 0.22.2

Citations

Cite Online DPO as:

@article{guo2024direct,
    title        = {{Direct Language Model Alignment from Online AI Feedback}},
    author       = {Shangmin Guo and Biao Zhang and Tianlin Liu and Tianqi Liu and Misha Khalman and Felipe Llinares and Alexandre Ram{'{e}} and Thomas Mesnard and Yao Zhao and Bilal Piot and Johan Ferret and Mathieu Blondel},
    year         = 2024,
    eprint       = {arXiv:2402.04792}
}

Cite TRL as:

@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}
}