If the abstract is accurate, then I'm very, very excited to try this on LLaMA 65B. We are tantalizingly close to ChatGPT performance parity on consumer hardware

If the abstract is accurate, then I'm very, very excited to try this on LLaMA 65B. We are tantalizingly close to ChatGPT performance parity on consumer hardware. Hopefully this lowers the cost of doing instruct fine tuning on the larger models, and we see a Vicuna like model based on LLaMA 65B soon. This is exciting folks.