mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2026-07-29 22:20:58 -05:00
d6a5094004
* Fix flashattention support check for devices that don't support subgroups * set path to none if kv_tile doesn't fit