---
library_name: openpi
tags:
- robotics
- openpi
- pi0
- franka
---

# GRASPNET_FINAL_h100

openpi π₀.₅ policy checkpoint.

- **Train config:** `pi05_Franka_GRASPNET_FINAL`
- **Experiment:** `GRASPNET_FINAL_h100`
- **Checkpoints in this repo:**

- `step_25000/`
- `step_49999/`

Each folder is one training step and is self-contained.
- **Training dataset:** [saifahmad123/GRASPNET_FINAL](https://huggingface.co/datasets/saifahmad123/GRASPNET_FINAL)

## Contents

| Path | Purpose |
|---|---|
| `step_49999/params/` | Model weights. Required to serve the policy. |
| `step_49999/assets/` | Normalization statistics. Required to serve the policy. |
| `step_49999/train_state/` | Optimizer state. Only present if uploaded with `--include-train-state`; needed to resume training. |

## Usage

```python
from huggingface_hub import snapshot_download
from openpi.policies import policy_config
from openpi.training import config as _config

ckpt = snapshot_download("saifahmad123/GRASPNET_FINAL_h100", allow_patterns="step_49999/*") + "/step_49999"
train_config = _config.get_config("pi05_Franka_GRASPNET_FINAL")
policy = policy_config.create_trained_policy(train_config, ckpt)

action_chunk = policy.infer(observation)["actions"]
```