R
EN 中
Back to projects
•
1 min read
Chinese-LLM LoRA Fine-tuning Framework

ChatGLM3 / Qwen / Baichuan2 / Yi · BLEU-ROUGE eval · Docker

  • Python
  • PyTorch
  • LoRA / QLoRA
  • PEFT
  • Gradio

A full pipeline from data preprocessing to deployment for efficiently fine-tuning mainstream Chinese LLMs.

  • Multi-model: ChatGLM3, Qwen, Baichuan2, Yi
  • Efficient training: LoRA + QLoRA quantized fine-tuning
  • Evaluation: built-in BLEU / ROUGE metrics
  • UI + tracking: Gradio compare interface, Weights & Biases experiment tracking, Docker one-click deploy