mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2026-07-26 12:40:58 -05:00
c9ced4910b
Run a preprocess to count how many times each expert is used, and use this to quickly discard workgroups that aren't needed.