https://stanford.io/3LMXpb4 - Alpaca: A Strong, Replicable Instruction-Following Model Alpaca, a new instruction-following language model, fine-tuned from Meta’s LLaMA 7B model, has been released for academic research purposes. It is trained on 52K instruction-following demonstrations generated using OpenAI’s text-davinci-003, and exhibits similar behavior to text-davinci-003 while being smaller and cheaper to reproduce. However, it also shows some common deficiencies of language models, including hallucination, toxicity, and stereotypes. The release includes an interactive demo, data, data generation process, and training code. While there are risks associated with the release, the benefits for the research community outweigh the risks. Future research opportunities include further evaluation, safety measures, and understanding the properties of base models and instruction data.
