Skip to content

Comment on State-of-the-Art Chatbot, Vicuna-7B, now runs on MacBook with GPU accelerationparent

Comments

Did they release the merged weights, yet? I'd love to try this model.

Afaict from the docs, you still need to request the original Llama weights from Meta (or get ahold of them another way), then apply the diff-weights requiring 60GB RAM?

You can find the merged weights pretty easily online, I wouldn't hold your breath waiting for an official release given the licensing issues around LLaMA.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.