metadata
license: apache-2.0
tags:
- lerobot
- smolvla
- vla-robotics
- so-arm101
library_name: lerobot
pipeline_tag: robotics
datasets:
- makepluscode/pick_red
ch10-01-smolvla-pick-red-60k
SmolVLA fine-tuned on makepluscode/pick_red for the SO-101 red-block pick task,
trained via LeLab for 60,000 steps (final loss ~0.084).
| Item | Value |
|---|---|
| Policy | SmolVLA |
| Base | lerobot/smolvla_base |
| Dataset | makepluscode/pick_red |
| Steps | 60,000 |
| Hardware | SO-101 leader/follower |
Load
from lerobot.policies.smolvla.modeling_smolvla import SmolVLAPolicy
policy = SmolVLAPolicy.from_pretrained("makepluscode/ch10-01-smolvla-pick-red-60k")
Pre/post-processors (policy_preprocessor.json / policy_postprocessor.json)
in this repo are loaded together by from_pretrained.