llama.cpp b10751 发布：CUDA 融合 MoE 加权归约优化

AI 摘要

llama.cpp 发布 b10751 版本，主要更新是在 CUDA 后端中融合了 MoE 加权专家归约操作，将多个内核合并为一个，以减少全局内存流量并提升性能。该融合支持未缩放和缩放两种图模式，并可通过环境变量禁用。同时发布了适用于多个平台和硬件的二进制文件。

正文节选

<details open> cuda: fuse MoE weighted expert reduction (#25952) * cuda : fuse MoE weighted reduction (mul + view + add) The MoE combine tail currently writes weighted expert outputs to global memory before reducing them. That intermediate global-memory traffic is the main cost. The production baseline generally runs two physical fused kernels; this path runs one. This change matches the full expert-weighting plus ordered-reduction subgraph and replaces it with one weighted-reduction kernel.

发布时间：2026-09-02 04:13

抓取时间：2026-09-02 04:19

来源机构：ggml-org