Serve poolside/Laguna-S-2.1-NVFP4 (117.6B MoE, 256K ctx, DFlash spec-decode) with vLLM on a single NVIDIA DGX Spark (GB10) — bare-metal, no sudo, systemd + watchdog, measured tuning recipe. Operated from a Mac over SSH.
By chatting or signing in you agree to the Terms and chat-message logging (revocable in History).