Pipeline-parallel LLM inference across GPUs on separate machines. - View it on GitHub