llama.cpp b10707 发布：优化 kv-cells 扫描提升长上下文生成速度

AI 摘要

llama.cpp 发布 b10707 版本，优化了 kv-cells 的序列扫描逻辑，在保持行为不变的前提下提前终止扫描。该优化显著提升了长上下文下的生成速度，例如在 RTX PRO 6000 上 55k 上下文从 56.3 t/s 提升至 74.3 t/s，132k 上下文从 33.6 t/s 提升至 50.9 t/s。

正文节选

<details open> kv-cells: stop the sequence scan once all sequences are seen (#28011) for_each_token_in tested all LLAMA_MAX_SEQ sequences for every used cell, while a cell almost always belongs to one. The scan now stops once the cell's own sequences have been seen. Same visit order, same callback arguments, so behaviour is unchanged. get_prev_tokens is the only caller, so this affects the n-gram path. RTX PRO 6000, Qwen3.8-Flash-Next UD-Q4_K_XL, fa on, warm runs: 55k context generatio

发布时间：2026-08-31 11:03

抓取时间：2026-08-31 11:42

来源机构：ggml-org