File size: 1,798 Bytes
537e2e3 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 | ---
library_name: diffusers
pipeline_tag: robotics
tags:
- robotics
- vision-language-action
- pi0.5
- diffusers
license: other
---
# PI0.5 Base — GigaModels PyTorch Conversion
This repository contains a PyTorch Diffusers-format conversion of the
Physical Intelligence PI0.5 base checkpoint. The weights were converted from
the official OpenPI JAX checkpoint with the following script from
[open-gigaai/giga-models](https://github.com/open-gigaai/giga-models):
- [`projects/vla/pi0/scripts/convert_jax_model_to_pytorch.py`](https://github.com/open-gigaai/giga-models/blob/main/projects/vla/pi0/scripts/convert_jax_model_to_pytorch.py)
- [GigaModels PI0/PI0.5 documentation](https://github.com/open-gigaai/giga-models/blob/main/projects/vla/pi0/README.md)
No additional fine-tuning was applied to these base weights. The tokenizer
files are colocated with the converted policy so consumers can use one Hub
repository for both the policy and tokenizer.
## Loading with GigaModels
```python
from giga_models import PI0Policy
policy = PI0Policy.from_pretrained("Miical/pi05-base")
```
## Loading with verl-vla
Set the model path directly to the Hub repository:
```yaml
cluster:
actor_rollout_ref:
model:
path: Miical/pi05-base
```
Normalization statistics are embodiment- and dataset-specific and are not
included in this base checkpoint. Configure `model.adapter.norm_stats_path`
when training or evaluating with verl-vla.
## Attribution and license
The conversion implementation is provided by GigaModels under Apache-2.0.
Use and redistribution of the converted weights remain subject to the terms
applicable to the original PI0.5 checkpoint and its underlying components.
Please also cite the original PI0/PI0.5 work as requested by the GigaModels
documentation.
|