ggrun

llama.cpp/ik_llama.cpp launcher: loads big MoE models across mismatched multi-GPU rigs by exact VRAM math.

GitHub

Official source

Start locally with the CLI guide