https://huggingface.co/mistralai/Mixtral-8x22B-Instruct-v0.1

#136
by yttria - opened

Mistral fixed the tokenizer

Thanks for notifying me, let's try to quant it at once :)

mradermacher changed discussion status to closed

Unfortunately, not even transformers can now load it without crashing. Maybe there is a special way to load it now, but that isn't supported by llama.cpp. I'll look around for a workaround, but it seems completely hosed now. Shouldn't have deleted it :()

Hmm, the configs of base and instruct differ. If I replace one by the other it seems to load. Hope that's correct...

That was a bit unclear, the base model didn't load, not the instruct model.

Sign up or log in to comment