llama.cpp b11866：Metal 移除 mul_mm_id amax 重缩放

AI 摘要

llama.cpp 发布 b11866 版本，主要变更在 Metal 后端：移除了 kernel_mul_mm_id 中针对 src1 的二次幂重缩放，改由计算图中的 per expert-token amax 缩放替代，从而在算子执行前将 MoE 下投影激活值保持在 f16 范围内。该改动删除了两阶段归约内核、其暂存缓冲区以及此前引入的 GGML_PREC_F32 调度和相关测试用例。发布同时提供 macOS、Linux、Windows、Android 等多平台预编译包。

正文节选

<details open> metal : remove mul_mm_id amax rescaling (#30924) The src1 power-of-two rescale in kernel_mul_mm_id is superseded by the per expert-token amax scaling done in the compute graph (ref: https://github.com/ggml-org/llama.cpp/pull/30861), which keeps the MoE down-projection activations within f16 range before the op runs. Drop the two-stage reduction kernels, their scratch buffer and the GGML_PREC_F32 dispatch added in https://github.com/ggml-org/llama.cpp/pull/26223, as well as the

发布时间：2026-10-25 03:51

抓取时间：2026-10-25 03:56

来源机构：ggml-org