llama.cpp b10353 发布：修复 CUDA/Metal 的 ROLL 操作

AI 摘要

llama.cpp 发布 b10353 版本，修复了 CUDA 和 Metal 后端中 ggml_roll 操作对非连续输入产生错误结果的问题。该修复要求 src 张量连续，并添加了相应的测试用例。同时提供了多个平台的预编译二进制文件。

正文节选

<details open> ggml : require contiguous src for ROLL on CUDA and Metal (#25928) ggml_roll only asserts nb[0] == ggml_type_size, so a permuted src is a valid input, but the CUDA and Metal roll kernels index by ne alone and never read the nb strides. A non-contiguous src therefore produced silently wrong results. Neither backend declared a contiguity requirement in supports_op, so the scheduler did not fall back to the CPU implementation, which does handle strides correctly. Add the requiremen

发布时间：2026-08-11 06:00

抓取时间：2026-08-11 07:04

来源机构：ggml-org