16 ms·Try the llama.cpp fork by thetom. It's called turboquant after the techniqueby beacon294 1mo agoTry the llama.cpp fork by thetom. It's called turboquant after the technique