CUDA: Hide latency of bias and gate-loading for fused `mul_mat_vec_q`#16847

am17an merged 1 commit intoggml-org:masterggml-org/llama.cpp:masterfrom

ORippler:osimons/prefetch_gate_bias_in_fused_mmvqORippler/llama.cpp:osimons/prefetch_gate_bias_in_fused_mmvqCopy head branch name to clipboard