llama.cpp b11630 修复 ggml conv_1d 批量输入布局错误

AI 摘要

llama.cpp 发布 b11630 版本，修复了 ggml 中 conv_1d 在批量输入下的结果布局错误：此前 GEMM 输出实际为 [OL, N, OC]，而 reshape 却声明为 [OL, OC, N]，导致通道与批次交错且无任何断言触发。修复后统一 reshape 为 [OC, N, OL] 并 permute 还原，所有批次大小构建相同计算图，新增测试覆盖七种几何配置。该修复对共享单一计算图的多路音频合成下游项目有实际意义。

正文节选

<details open> ggml: fix conv_1d result layout for batched input (#28738) * ggml: fix conv_1d result layout for batched input The GEMM behind conv_1d folds the batch into its rows, so the product sits in memory as [OL, N, OC] while the reshape claims [OL, OC, N]. The shape is right and the contents are transposed, so nothing fires: no assert, no abort, the graph just runs and the channels come out interleaved with the batches. With N == 1 the two readings describe the same buffer, which is wh

发布时间：2026-10-15 08:32

抓取时间：2026-10-15 09:01

来源机构：ggml-org