This is a heavily interactive web application, and JavaScript is required. Simple HTML interfaces are possible, but that is not what this is.
Post
lhl
lhl.bsky.social
did:plc:4tvzofz76boqj3giiwtepeiz
Over the weekend, I finished up our Llama 405B run (4th group I know of to do a FFT?). It was a real beast to train, but beats our Shisa V2 70B (as well as GPT-4 and GPT-4 Turbo) using basically our Shisa V2 recipe. It is, I believe the best performing LLM (JA and EN) to ever be trained in Japan.
2025-04-28T12:25:10.379Z