You can deploy this model and start earning money today!

Miical/
pi05-libero

Name: pi05-libero
Availability: InStock

diffusers

safetensors

robotics

vision-language-action

pi0.5

libero

other

Views

No views yet

PI0.5 LIBERO — OpenPI JAX Checkpoint Conversion

This repository contains a PyTorch Diffusers-format conversion of the PI0.5 LIBERO JAX checkpoint provided by the official Physical Intelligence OpenPI repository. The weights in this repository were not fine-tuned by verl-vla.

The OpenPI JAX checkpoint was converted with the following implementation from open-gigaai/giga-models:

The converted checkpoint is stored in the native Diffusers PI0Policy layout. Tokenizer files are colocated with the 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-libero")

Loading with verl-vla

Set the model path directly to the Hub repository and point the LIBERO embodiment adapter to the included normalization statistics:

yaml

cluster:
  actor_rollout_ref:
    model:
      path: Miical/pi05-libero
      adapter:
        embodiment: libero
        norm_stats_path: /path/to/norm_stats.json

Normalization statistics

norm_stats.json contains the state and action normalization values associated with the converted OpenPI checkpoint. The values were extracted unchanged from the source checkpoint configuration and moved into a separate artifact so config.json only describes the policy architecture.

Download norm_stats.json together with the checkpoint and pass its local path through model.adapter.norm_stats_path when using the policy with verl-vla.

Attribution and license

The source checkpoint is provided by the Physical Intelligence OpenPI project. The PyTorch 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 follow the citation and attribution guidance from OpenPI and GigaModels.