llama: re-create the KV cache when flash attention resolves to disabled (performance and layout bug)#26460

wanghqc wants to merge 1 commit into

ggml-org:masterggml-org/llama.cpp:masterfrom

qualcomm:hq/llama-fa-auto-recreate-kv-cachequalcomm/llama.cpp:hq/llama-fa-auto-recreate-kv-cacheCopy head branch name to clipboard