--- tags: - robotics - vision-language-action - flower-vla - lerobot - robotis-sg2 --- # Task 000759 FLOWER VLA - B200 - step 5000 FLOWER VLA fine-tuned on the ROBOTIS SG2 Task 000759 LeRobot v3 dataset. ## Training contract - Upstream: `intuitive-robots/flower_vla_calvin` at `acb32ee85719e51aa94139b184c64b1b29f11d33` - GPU: NVIDIA B200 MIG 3g.90gb - Precision: BF16 mixed precision - Batch size: 8 - Checkpoint step: 5,000 of 10,000 - Optimizer: AdamW - Learning rate: 2e-5 - State/action: 22D / 22D - Action horizon: 10 - Cameras: left head, left wrist, right wrist - Dataset: 98 episodes, 10,578 frames - Validation and robot rollout: disabled during training - Base weights: official FLOWER 360000-step checkpoint The SG2 action/proprioception heads and the six DiT layers absent from the public 12-layer checkpoint are initialized for this fine-tuning setup. The compatible Florence-2 and FLOWER weights are retained. This repository contains a full PyTorch Lightning checkpoint, including model, EMA, optimizer, scheduler, and global-step state. It is intended for offline evaluation and controlled inference integration. It does not authorize robot actuation. Offline MAE will be added after the uploaded checkpoint is downloaded and evaluated with the fixed validation split (5%, seed 242).