Hacker News new | ask | show | jobs
by wirybeige 104 days ago
The vulkan backend for llama.cpp isn't that far behind rocm for pp and tp speeds