llama.cpp b11440 修复 NextN 提取调度器预留问题

AI 摘要

llama.cpp 发布 b11440 版本，修复了推测解码 MTP 初始化中 NextN 提取标志变化后调度器未重新预留的问题。当未掩码提取启用时，主干图保留所有 token 导致首次解码按批次形状重新分配，后续更宽批次会触发 GGML_SCHED_DEBUG_REALLOC 错误。修复方式是在标志变化时使预留失效，以便下次计算按新图形形状重新预留。该版本同时提供 macOS、Linux、Windows、Android 等多平台预编译二进制。

正文节选

<details open> llama : re-reserve the sched when the nextn extraction flags change (#30020) The speculative MTP init enables NextN extraction on the target and draft contexts after both were created and their schedulers reserved. With unmasked extraction the trunk graph keeps every token through the last layer instead of cropping to the output rows, so the first decode reallocates to that batch's shape and the next, wider batch trips GGML_SCHED_DEBUG_REALLOC. Invalidate the reserve when the fl

发布时间：2026-10-06 21:07

抓取时间：2026-10-06 22:12

来源机构：ggml-org