mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2026-09-09 06:19:04 +02:00
d6a5094004
* Fix flashattention support check for devices that don't support subgroups * set path to none if kv_tile doesn't fit