Llama3.2-1B-DuDi

Model Card for Llama3.2-1B-DuDi

This model is a fine-tuned version of Llama-3.2-1B. 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="aisingapore/Llama3.2-1B-DuDi", 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 DuDi, a method introduced in DuDi: Dual-Signal Distillation with Cross-Lingual Verbalizer.

Framework versions

• TRL: 0.29.0
• Transformers: 4.57.6
• Pytorch: 2.9.1+cu129
• Datasets: 4.7.0
• Tokenizers: 0.22.2

Citations

Cite DuDi as:

@misc{payoungkhamdee2026dudidualsignaldistillationcrosslingual,
      title={DuDi: Dual-Signal Distillation with Cross-Lingual Verbalizer}, 
      author={Patomporn Payoungkhamdee and Tinnakit Udsa and Jian Gang Ngui and Sarana Nutanong and Alham Fikri Aji and Peerat Limkonchotiwat},
      year={2026},
      eprint={2606.04694},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2606.04694}, 
}

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}
}