6.4x faster than llama.cpp, 3.9x faster than MLXbasecompute.co4 points·lukasonedge··2 commentsOpen articleSaveView on HN