Downloads · 30 days
32
37% of all-time downloads
pi0-libero-pytorch
Fisher-Wang/pi0-libero-pytorch
pi0-libero-pytorch is a machine learning model from Fisher-Wang. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
PyTorch conversion of Physical Intelligence's official pi0libero checkpoint (gs://openpi-assets/checkpoints/pi0libero, π₀ fine-tuned on the LIBERO benchmark), converted via openpi examples/convertjaxmodeltopytorch.py…
All-time downloads
87
Public
Parameters
3.5B
14 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors14 GB · 100%
At a glance
- License
- apache-2.0
- Access
- Public
- Created
- Jul 12, 2026
- Updated
- Jul 12, 2026
- SHA
- 23d60043
Model card
From the Hugging Face model README
pi0-libero-pytorch
PyTorch conversion of Physical Intelligence's official pi0_libero checkpoint
(gs://openpi-assets/checkpoints/pi0_libero, π₀ fine-tuned on the LIBERO
benchmark), converted via openpi examples/convert_jax_model_to_pytorch.py
with --config_name pi0_libero, precision float32. Uploaded for RobotTrain
subnet validator pipeline testing.
Unlike the DROID/base negative-control repos in this collection, this is a genuine LIBERO policy: it was fine-tuned on LIBERO with the LIBERO IO convention and ships its own norm_stats. As a previous-generation model (π₀ vs π₀.₅) it provides a mid-range positive sample for benchmark discrimination tests.
Layout: model.safetensors (fp32) + assets/<asset_id>/norm_stats.json
(original, z-score stats — π₀ does not use quantile normalization).
Load with openpi: create_trained_policy(get_config("pi0_libero"), <this dir>).
- License
- apache-2.0
- Created
- Jul 12, 2026
- Updated
- Jul 12, 2026