For LLM workloads this might help? https://docs.tinygrad.org/tinygpu/
Interesting, I've never seen this, thanks