> ## Documentation Index
> Fetch the complete documentation index at: https://docs.orchestrallabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# OpenPI checkpoint imports

> Create a Hugging Face repository for Servo Pi0.5 imports.

Put one checkpoint at the Hugging Face repository root:

* **JAX/Orbax:** contents of one [OpenPI training step](https://github.com/Physical-Intelligence/openpi/blob/main/src/openpi/training/checkpoints.py): `params/` and exactly one `assets/<id>/norm_stats.json`. For non-default inference settings, add [`openpi_config.json`](/schemas/openpi_config.schema.json) at the root.
* **OpenPI PyTorch:** contents of the [conversion output](https://github.com/Physical-Intelligence/openpi/blob/main/examples/convert_jax_model_to_pytorch.py): `config.json`, `model.safetensors`, and `assets/<id>/norm_stats.json`. Import with `--openpi-config-name NAME` using a config registered by Servo.
* **LeRobot:** contents of one checkpoint's `pretrained_model/` directory, following [LeRobot's checkpoint layout](https://github.com/huggingface/lerobot/blob/main/docs/source/multi_gpu_training.mdx): `config.json`, `model.safetensors`, `policy_preprocessor.json`, `policy_preprocessor_step_3_normalizer_processor.safetensors`, `policy_postprocessor.json`, and `policy_postprocessor_step_0_unnormalizer_processor.safetensors`.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.