[CUDA] Increase number of output elements per-thread block if the K-dimension is small#20635

gaugarg-nv wants to merge 3 commits intoggml-org:masterggml-org/llama.cpp:masterfrom

gaugarg-nv:small_k_optimizationgaugarg-nv/llama.cpp:small_k_optimizationCopy head branch name to clipboard