No, a GPU would be ideal but not everyone has one, especially one with enough VRAM. I have an AMD card with 12gb of VRAM and I can run 7B - 13B models but even the 7B models (which seems to be the lowest that is still good) use a little more than 8gb of VRAM and most people probably have an Nvidia card with 8gb or less. 13B models get very close to using the full 12gb.
Is running using a GPU a bad thing? I’m new to this…
No, a GPU would be ideal but not everyone has one, especially one with enough VRAM. I have an AMD card with 12gb of VRAM and I can run 7B - 13B models but even the 7B models (which seems to be the lowest that is still good) use a little more than 8gb of VRAM and most people probably have an Nvidia card with 8gb or less. 13B models get very close to using the full 12gb.
Not everyone has a dedicated GPU, I would guess. GPUs are good at doing tensor calculations, but they’re not the only way.
Its better if you have a good GPU, but will not run without a card from the last few years. It can run on CPU but it’s much slower.