llama.cpp b12189 修复 Metal GROUP_NORM 批次与非连续输入问题

AI 摘要

llama.cpp 发布 b12189 版本，修复 Metal 后端 GROUP_NORM 算子在 ne03>1 及非连续输入下的错误：原内核忽略批次维度，导致除首批外数据未被写入。修复后按 (group, i03) 分派线程组、按批次计算统计量，并支持跨步访问与 F32 类型门控。该更新提升 Apple Silicon 上相关算子的正确性，同时提供各平台预编译包。

正文节选

<details open> metal : fix GROUP_NORM for ne03 > 1 and non-contiguous inputs (#31852) The kernel_group_norm_f32 kernel and its host dispatch ignored the batch dimension (ne03) of the source tensor: only ngrp threadgroups were dispatched and each normalized the flat range [tgpig*gs, (tgpig+1)*gs) with gs = ne00*ne01*ceil(ne02/ngrp). For tensors with ne03 > 1, all but the first ne00*ne01*ne02 elements were left unwritten (exposed by the new GROUP_NORM test with ne=[3,1,1,65537], num_groups=1).

发布时间：2026-11-07 20:15

抓取时间：2026-11-07 21:22

来源机构：ggml-org