llama_moe_cache: run missed experts on CPU, add tuning script by am17an · Pull Request #30966 · ggml-org/llama.cpp
retained text
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.
document fetched 2026-11-08 00:34 · canonical https://reddit.com/r/LocalLLaMA/comments/1xqacka/llama_moe_cache_run_missed_experts_on_cpu_add