Comment by bensyverson

I do think these “run a bigger model than will fit in VRAM” projects are necessary steps, but are they functionally useful or helpful to anyone currently? For example, is anyone out there running a big Qwen for coding on a 16-32GB machine with these techniques?