This model is amazing.
#1
by
boeroboy
- opened
I'm blown away by both the knowledge and the speed of this model after using it casually for a few weeks now. It runs amazing on just about any device and fits in just under 8GB of RAM. It's slightly too big for any 8GB GPU I throw at it but I've also used it in large context CPU samples (100,000+ tokens) and it performs beautifully within ~80GB. Cheers to you magnificent trainers.
What original model? The files are still here. Do you mean original Amethyst models?
Well i did say original.. so that would imply i'm not talking about Bloke's conversions
https://huggingface.co/Undi95/Amethyst-13B-Mistral = 404.