llama.cpp b11879：CUDA MMQ 融合 FFN gate/up 与 GLU

AI 摘要

llama.cpp 发布 b11879 版本，在 CUDA 后端为 MMQ 量化矩阵乘实现 FFN gate/up 矩阵乘与 GLU 的融合：通过将 gate 与 up 权重行按 8 行交错排布，使一次 mul_mat_q 启动即可完成两个矩阵乘并在写回时计算 GLU。该优化面向 prompt 处理阶段，支持 Q4_K、Q5_K、Q6_K、Q8_0 和 IQ4_XS 量化类型，适用于 NVIDIA Turing 及更新架构。

正文节选

<details open> CUDA: fuse FFN gate/up matmuls and GLU in MMQ with interleaved gate/up rows (#29948) * cuda: fuse FFN gate/up and GLU in MMQ with interleaved gate/up rows For prompt processing, run the gate and up matmuls of a dense FFN as one mul_mat_q launch and apply the GLU in the write-back. - The loaders take the gate weights as a second pointer and fill the weight tile with up and gate rows interleaved in chunks of 8, so each thread holds an up value and the matching gate value (ro

发布时间：2026-10-26 00:12

抓取时间：2026-10-26 00:59

来源机构：ggml-org