[belval]

Nice project! I could not find the information in the README.md, can I run this with a GPU? If so what do I need to change? Seems like it's hardcoded to 0 in the run script: https://github.com/getumbrel/llama-gpt/blob/master/api/run.s...

[dicriseg]

I put up a draft PR to demo how to run it on a GPU: https://github.com/getumbrel/llama-gpt/pull/11

It breaks other things like model downloading, but once I got it to a working state for myself, I figured why not put it up there in case its useful. If I have time, I'll try to rework it a little bit with more parameters and less dockerfile repetition to fit the main project better.

[mayankchhabra]

Ah yes, running on GPU isn't supported at the moment. But CUDA (for Nvidia GPUs) and Metal support is on the roadmap!

[samspenc]

Ah fascinating, just curious, what's the technical blocker? I thought most of the Llama models were optimized to run on GPUs?

[mayankchhabra]

It's fairly straightforward to add GPU support when running on the host, but LlamaGPT runs inside a Docker container, and that's where it gets a bit challenging.

[stavros]

It shouldn't, nVidia provides a CUDA Docker plugin that lets you expose your GPU to the container, and it works quite well.

[dicriseg]

See above if you're interested in that. It does work quite well, even with nested virtualization (WSL2).

[stavros]

I am, thanks!

[crudgen]

Had the same thought, since it is kinda slow (only have 4 pyhsical/8 logical cores though). But I think vRAM might be a problem (8gb can work, if one has a rather recent gpu (here m1/2 might be interesting)).