llama.cpp b11936：k-pool 稀疏选择与 kq_mask 形状解耦
原标题：b11936
AI 摘要
llama.cpp 发布 b11936 版本，核心改动是将 k-pool 稀疏选择（sparse selection）的构建与 kq_mask 的形状解耦：qwen4exp（QSA）和 glm5-next 按索引选取缓存单元，不再读取 kq_mask 值，选择张量改为基于缓存单元的 [n_kv, n_tokens] 布局，并在注意力计算前拆分为 K 视图的流。该改动使稠密 mask 可以改变布局（如按序列的行列表）而不影响稀疏层，同时清理了未使用的 build_fused_top_k_mask 参数。
正文节选
<details open> llama : build the k-pool sparse selection independently of the kq_mask shape (#31243) * llama : build the k-pool sparse selection independently of the kq_mask shape qwen4exp (QSA) and glm5-next pick cache cells by index and no longer read the kq_mask values, but they still shaped the selection like the dense kq_mask. Build the selection as [n_kv, n_tokens] over the cache cells instead, and split it into the streams of the K view right before the attention. This lets the dense