Proposed architecture for inferencing sparse MOE models increasing Active parameters using layered + linear decay. Succinct reasoning without any model training or fine tune. [p]

I ported MoE expert expansion to llama.cpp 🚀 Run MoE models with MORE routed experts than the native top-K (8->x), adaptive threshold, 99→50% influence decay, layer range. Runtime-only, all backends. Tested on Qwen 3.6 35B A4B+ https://github.com/vagrillo/llama.cpp/blob/moe-expansion/docs/moe-expans