DGX Spark Triple Decker setup

I’m sharing scripts for anyone interested in setting up a 2- or 3-node Spark cluster. GGUF-focused. In a nutshell — it shares your LM Studio model library across all nodes (via NFS), full native RDMA support over ConnectX-7, and an OpenAI-compatible API endpoint on the head node. Tested with opencode against MiniMax-M2.7 Q6, including a 20-minute codebase architecture audit that exercises tool calling pretty heavily.