Enhance CUDA flash attention kernel selection for DKQ=512 with low gq…#6

Ooooze wants to merge 1 commit into

feature/turboquant-kv-cacheAtomicBot-ai/atomic-llama-cpp-turboquant:feature/turboquant-kv-cachefrom

fix/cuda-mma-dkq512-fallbackAtomicBot-ai/atomic-llama-cpp-turboquant:fix/cuda-mma-dkq512-fallbackCopy head branch name to clipboard