Show HN: Llama.cpp fork with 2-4x multiGPU speed for MoE models bigger than VRAM

(github.com) by neuralll | Sep 25, 2026 | 0 comments on HN