rebot-smolvla-pickthebreadintothepot-valid4-42eps-20k
SmolVLA checkpoint trained for the reBot hackathon validation stack.
- Action:
pickthebreadintothepot
- Training steps:
20000
- Dataset:
phi-media-lab/rebot_pickthebreadintothepot_20260425_42eps_valid4
- Source checkpoint path on Ali L20:
outputs/train/rebot_smolvla_pickbread_pot_42eps_valid4_20260425_20000steps/checkpoints/010000/pretrained_model
- Video backend used for training:
pyav
- Batch size:
8
- Validation: CUDA load +
policy.select_action smoke test passed, output action shape (1, 7).
This repo contains the LeRobot pretrained_model folder contents and can be loaded with policy.path=fbsh96/rebot-smolvla-pickthebreadintothepot-valid4-42eps-20k or downloaded as a local pretrained model.
