llama.cpp b12101：GGUF 写入器直接写 host-buffer 张量数据

AI 摘要

llama.cpp 发布 b12101 版本，核心改动在 ggml 的 GGUF 写入器：对 host-buffer 张量跳过临时向量暂存，直接将数据写入文件，省去逐张量的内存分配、零初始化和 memcpy。该版本同时提供 macOS、Linux、Windows、Android 等多平台及 CUDA、Vulkan、ROCm、SYCL 等后端构建产物。改动可降低模型转换与 GGUF 写出时的内存开销和拷贝成本。

正文节选

<details open> ggml : write host-buffer tensor data directly in gguf writer (#31144) The gguf writer staged each tensor into a temporary vector before writing it. For host-buffer tensors the data pointer is directly writable, so skip the staging copy and write it straight to the file. This avoids the per-tensor allocation, zero-init and memcpy. Assisted-by: dsh:llama.cpp/ggml-org/DeepSeek-V4-Flash-Vision-Exp-GGUF (Mac Studio, M2 Ultra, 192GB) </details> **Website:** - <https://llama.app> *

发布时间：2026-11-04 16:32

抓取时间：2026-11-04 16:52

来源机构：ggml-org