TY  - RPRT
TI  - Efficient Training of Robust Traditional Chinese LLaMA-1B on a Single Consumer GPU: Continual Pre-training, SFT, and DPO
AU  - Yu-Cheng Chih
AU  - Ming-Tao Duan
AU  - Yong-Hao Hou
PY  - 2025
DO  - 10.5281/zenodo.17240351
UR  - https://arxiv.org/abs/2510.01616
ID  - 2510.01616
ER  -