llama.cpp b11227：修复 causal_attn 切换导致的调度器重复预留

AI 摘要

llama.cpp 发布 b11227 版本，核心改动是修复 llama_context::set_causal_attn() 在切换 causal_attn 标志时每次都触发调度器完整重新预留（sched_reserve）的问题。由于 Gemma 模型处理视觉输入时每个非因果图像块会两次翻转该标志，导致每张图片产生两次昂贵的重新预留，多图或视频输入尤为缓慢。修复后 H200 与 RTX 4090 上多图场景延迟最高分别降低约 3.4 倍和 7.6 倍，生成输出保持不变。

正文节选

<details open> context : do not re-reserve the scheduler when toggling causal_attn (#28751) * context : do not re-reserve the scheduler when toggling causal_attn `llama_context::set_causal_attn()` marks the scheduler to do a full re-reserve on every change of the flag. For vision inputs, this flag is flipped twice around each non-causal image chunk for Gemma models, resulting in two expensive `sched_reserve()` passes per image. This is especially slow for multi-image or video inputs. The cos

发布时间：2026-09-28 17:29

抓取时间：2026-09-28 17:39

来源机构：ggml-org