A simple, unified multimodal models training engine. Lean, flexible, and built for hacking at scale.
Terminal score: 0–1000 raw, weighted across 4 dimensions. Public score: 0–10 normalized (shown in the 30-day stars chart above).
