2024-03-14

This is a 3-bit exl2 (Exllamav2) conversion of MistralAI's Mixtral-8x7B-v0.1

This takes about 17GB on disk, so it should load on consumer cards with 24GB of VRAM.

More on ExLlamav2 -> https://github.com/turboderp/exllamav2 ( works with Oobabooga's textgen web ui )
Original Model -> https://huggingface.co/mistralai/Mixtral-8x7B-v0.1 ( Original files are ~87GB on disk )

Model is too large to load in Inference API (serverless). To try the model, launch it on Inference Endpoints (dedicated) instead.