cuda : prevent integer truncation and overflow errors when using KQ mask strides in flash_attn_mask_to_KV_max kernel#24945

fairydreaming wants to merge 1 commit into

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

fairydreaming:stride-narrow-conv-fixfairydreaming/llama.cpp:stride-narrow-conv-fixCopy head branch name to clipboard