RoboColosseum/yam-microwave-molmoact2

MolmoAct2 — BimanualYAM Microwave

• Task: "Open the microwave and take out the bowl." — 100 episodes, 108,487 frames @ 30 fps (RoboColosseum/BimanualYAM-datasets/Microwave)
• Inputs/outputs: cameras top / left wrist / right wrist + instruction; 14-D absolute joint state and action
• Base: allenai/MolmoAct2 (base)
• Trained: Full fine-tune (VLM + action expert)
• Optim: LLM LR 2e-5, ViT/connector 1e-5, action expert 1e-4 (recipe x 128/64), warmup, decay to 0.1x; GBS 128 (32/GPU x 4); 5 epochs; fp32 master weights, bf16 compute; 4x H100; wall time 8h50m
• Files: Final checkpoint converted to HF format (model-*.safetensors, processor, norm_stats.json).
• Load: HF trust_remote_code model; see inference.py in this folder
• Code: RoboColosseum harness (scripts/train/molmoact2.sh); logs in W&B project RoboColosseum