llama.cpp b11284：修复 OpenVINO 量化权重视图 GET_ROWS

AI 摘要

llama.cpp 发布 b11284 版本，主要修复 OpenVINO 后端对量化权重视图的处理：在收集权重常量时解析 view_src，使量化权重的视图不再变成动态类型参数，并将视图的行偏移折叠进 gather 索引，而非切分反量化子图。该改动同时解除了对带非零偏移的量化 src0 视图的 supports_op 拒绝，使 #28253 中的 vs0 GET_ROWS 用例可在 OpenVINO 上运行。版本附带 macOS、Linux、Windows、Android 等多平台预编译包。

正文节选

<details open> openvino: serve GET_ROWS on a weight view from the base Constant (#28381) * openvino: serve GET_ROWS on a weight view from the base Constant Resolve view_src when collecting weight Constants so a view over a quantized weight no longer becomes a dynamic typed Parameter, and fold the row offset of the view into the gather indices instead of slicing the dequantization subgraph. * openvino: lift the quantized GET_ROWS view rejection The supports_op rejection of a quantized src0 v

发布时间：2026-09-30 20:35

抓取时间：2026-09-30 21:11

来源机构：ggml-org