2026-11-07 · America/Los_Angeles · 社区动态 · #1

llama_moe_cache: run missed experts on CPU, add tuning script by am17an · Pull Request #30966 · ggml-org/llama.cpp

This is a significant speedup for MoE offloading (low VRAM people). I tried Strata once, and it had some tuning mechanism, so probably this is a similar idea.

热度 51.3 / 100；排名与评分保留该期记录。