mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2026-07-03 01:30:23 +00:00
fadde67135
* Single load for half2 * Store scales in local mem * Vec load quantized values