Comment by DrNosferatu

Sounds great!

How far is it actually compatible right now?

Are there any tests / benchmarks?

Can this be used to run CUDA-accelerated LLMs?