Dev builds llama.cpp fork enabling MoE expert expansion

A developer created a llama.cpp fork implementing training-free MoE expert expansion, activating more routing experts than the native top-K at runtime, tested with GLM models.

2026-09-07 ~ 2026-09-07 · 2 related posts

• Custom llama.cpp Branch Adds Expert Expansion for MoE Models — Specific-Tax-6700 · 2026-09-07
• MoE expert expansion ported to llama.cpp: route more experts than native top-K at runtime, no training — Specific-Tax-6700 · 2026-09-07