Alpaca-LoRA
lastmod 2023-03-24created 2023-03-241 edits
Low-Rank LLaMA Instruct-Tuning
Features
- Instruction-tuning LLaMA on personal hardware using Low-Rank Adaptation
- A model of quality similar to text-davinci-003 that can still run on a Raspberry Pi
- The training code runs within 5 hours on a single RTX 4090
- Uses Hugging Face's PEFT (Parameter-Efficient Fine-Tuning) and Tim Dettmer's bitsandbytes for cheap and efficient fine-tuning