Mistral-7B-Instruct-v0.2 - I'm using this exact model too, and it is mind blowing, but to get the most out of it, make sure you use llama.cpp and turn on self-extend (I'm not sure if support for self-extend has been merged into main yet, I manually merged a dev branch)

I asked this another place in this thread, but curious to know how you plan to run and deploy this model. What's the cheapest or I guess most cost efficient way to do this without burning money. I'm a college student for reference so I don't have a lot of money lying around to experiment.

You’re right it’s expensive,

I have infra in my house, not gonna lie it cost a lot, I have a rack with 30k of equipment in it (including 5 GPUs)

But this would probably run on an AWS P2 instance with is 0.90 USD an hour, or there’s lambda labs which is also pretty cheap (no affiliation, just satisfied customer)

Thanks for the info! Appreciate it

What’s the cheapest machine I could Buy to do this?

Mistral-7B will run on a Raspberry Pi at q04 and a little bit of patience. For good acceleration though, you'll want an Nvidia machine with enough VRAM to comfortable store the model.

Shooting from the hip here I’d say $800 would be a good start for all new parts. You might get down to $500 with used parts.

The most expensive piece in either build is a video card. You want to be able to load the LLM file into your video card RAM.

I just got a 16gb new card for this. I can load up to 34b models on it but poorly. Anything 13b or less runs perfectly. A 12gb card would be able to run 7b models and with the right training I think 7b models can be awesome.

New 4060ti Used 3090

What's the cheapest way to run and deploy a fine-tuned model today? If I understand correctly I'll need to run cloud GPUs right

+1 interested to know this as well.

Mistral-7B-Instruct-v0.2 is amazing. I've set it as the default model of RecurseChat (a local AI chat app).

It works great, until recently WizardLM 2 came out and I tried it with local RAG. The difference is quite significant: https://twitter.com/chxy/status/1780101542311579942

Considering switching the default model after they put it back on HF.