LLM throughput
Live token/s from /validate/ping (Paddler seats, not 1:1 GPU pins).
Quick links to Forge, inference workers, admin consoles, and ops tools. Green dots = reachable from this browser (public URLs only). LAN links work on your home network or Tailscale.
Per-GPU 1:1 model binding UI is deprecated — use Paddler presets via Forge Routes / Mission Control. Wall GPU dials remain read-only.
Live token/s from /validate/ping (Paddler seats, not 1:1 GPU pins).
Loading…
Runs the Native Rust Sentinel-2 Turbidity & Curvelet Line-Snapping pipeline on all pending high-priority clusters.
Click to run on cesarops2. P100 down unloads nvidia (quiet idle). iLO fan curve needs patched iLO 2.77 + physical maintenance DIP switch for first flash.
Fires up the NIM Llama-90B vision classifier on cesarops3 and queues batches. Results feed the MAN_MADE review (served on T440). Defaults match current settings: 10 batches × 150 tiles.
Node IPs auto-discovered via fleet-hq registry (/fleet/v1/fleet). No hard-coded IPs.