CUDA: use MMVF for thin f16/bf16 mul_mat at small batch size - #29633

CUDA: use MMVF for thin f16/bf16 mul_mat at small batch size#29633

ynankani merged 2 commits into

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

ynankani:ynankani/cuda_bf16_mmvf_small_batchynankani/llama.cpp:ynankani/cuda_bf16_mmvf_small_batchCopy head branch name to clipboard