Hacker News Reader

If you can run any models on llama.cpp, that might be a good indicator of which StableLM models you'll be able to run.

I easily ran 7B int 4 ggml models on an MBP with 16gig RAM. Same works on a MBA with 8 gig RAM, but you'll have to not run any other memory-hogging app.

In 4bit 7B requires 6GB of RAM and runs at ChatGPT speeds on CPU (with llama.cpp).

The 15B model coming out soon will require 12GB of RAM and still run at good speeds on CPU.

On top of what others said, unlike SD, its not unusable on CPU... just very slow.

Stable diffusion will run on a 4GB GPU though.

The tuned 7B model is around 33 GBs, so you'll need a PC with that much VRAM or RAM. I haven't tried to load it on text generation ui though.