Instructions to use abcmaster123/pi0-record-testv2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- LeRobot
How to use abcmaster123/pi0-record-testv2 with LeRobot:
- Notebooks
- Google Colab
- Kaggle
| library_name: lerobot | |
| license: apache-2.0 | |
| base_model: lerobot/pi0_base | |
| tags: | |
| - lerobot | |
| - pi0 | |
| - robotics | |
| - vla | |
| - so101 | |
| datasets: | |
| - Ratstar216/record-testv2 | |
| pipeline_tag: robotics | |
| # pi0 — record-testv2 (SO-101) | |
| [LeRobot](https://github.com/huggingface/lerobot) の **π0 (pi0)** を | |
| [`Ratstar216/record-testv2`](https://huggingface.co/datasets/Ratstar216/record-testv2) | |
| で fine-tune したポリシーです(`lerobot/pi0_base` からの微調整)。 | |
| ## Training | |
| | | | | |
| |---|---| | |
| | Base model | `lerobot/pi0_base` | | |
| | Dataset | `Ratstar216/record-testv2`(43 episodes / 22,878 frames, LeRobot v3.0, SO-101 / so_follower) | | |
| | Cameras | `observation.images.front`, `observation.images.upper`, `observation.images.depth`(各 480×640) | | |
| | State / Action | 6-dim(shoulder_pan/lift, elbow_flex, wrist_flex/roll, gripper) | | |
| | Steps | 6000(batch size 4, ≈1.05 epoch) | | |
| | Precision | bfloat16, gradient checkpointing, `train_expert_only=true` | | |
| | Optimizer | AdamW, lr 2.5e-5, cosine decay + warmup(1000) | | |
| | Hardware | NVIDIA GH200 (Miyabi-G) | | |
| このリポジトリは step 6000 のチェックポイントです。 | |
| ## Usage | |
| ```python | |
| from lerobot.policies.pi0.modeling_pi0 import PI0Policy | |
| policy = PI0Policy.from_pretrained("abcmaster123/pi0-record-testv2") | |
| ``` | |
| > 注: LeRobot の π0 実装で読み込んでください(`lerobot` は π0 対応版が必要)。 | |