pi0.5 SO101 Stacking Big Rings (Isambard)
Fine-tuned pi0.5 checkpoint for SO101 big-ring stacking. Trained on Isambard GH200 (4x GPU node). Training is in progress.
Published checkpoints
| Directory | Step | Notes |
|---|
| step_15000/params/ | 15,000 | Interim |
Config
- Config name: pi05_so101_stacking_big_rings
- Experiment name: so101_stacking_big_rings
- Model: pi0.5 (pi05=True, action_horizon=30)
- Batch size: 32, LR: 2.5e-5 cosine (1k warmup), EMA: 0.999
- Delta actions: yes, Default prompt: stack the big rings
Dataset
- lorenzouttini/so101_stacking_big_rings
Usage
from openpi.training.config import get_config
config = get_config("pi05_so101_stacking_big_rings")
