GR00T-N1.7-3B finetuned on the pick skill, 403 train episodes (22 held out, deterministic split). 10000 steps, global batch 40, lr 1e-4. Includes processor/ and experiment_cfg/ for inference.
Files info
