Facico/Chinese-Vicuna
A Chinese instruction-following LLaMA-based model fine-tuned with LoRA for low-resource training on consumer GPUs.
Velocity · 7d
+3.5
★ / day
Trend
→steady
star history
This project provides methods to build and share Chinese instruction-following models based on LLaMA using LoRA parameter-efficient fine-tuning. It enables training Llama-7B instruction models on a single RTX-2080Ti (11GB) and multi-round chatbots with 2048 context length on an RTX-3090. The approach aims for high parameter efficiency, GPU friendliness, and easy deployment while following the Alpaca/Vicuna development paradigm.