2026-10-23 · America/Los_Angeles · 社区动态 · #20

qwen4exp: build the QSA selection without the kq_mask by sanmai · Pull Request #30965 · ggml-org/llama.cpp

gigabytes of VRAM saved (Qwen Flash Next) "This is one of the things Strata does to squeeze the most juice from the experts' cache."

热度 47.1 / 100；排名与评分保留该期记录。