fine-tuning-serving-openpi

# OpenPI Fine-Tuning and Serving

End-to-end workflows for fine-tuning and serving Physical Intelligence's OpenPI models (pi0, pi0-fast, pi0.5) on robot manipulation tasks from the public `openpi` repository. Covers blank-machine setup, JAX training, PyTorch training, checkpoint conversion, and policy inference serving.

## Quick start

Clone the public repo, install the workspace, then serve a pretrained policy:

```bash
git clone --recurse-submodules https://github.com/Physical-Intelligence/openpi.git
cd openpi
GIT_LFS_SKIP_SMUDGE=1 uv sync
GIT_LFS_SKIP_SMUDGE=1 uv pip install -e .
uv run scripts/serve_policy.py --env DROID
```

```python
from openpi_client import websocket_client_policy

client = websocket_client_policy.WebsocketClientPolicy(host="localhost", port=8000)
result = client.infer(observation)
actions = result["actions"]  # numpy array of shape (chunk_size, action_dim)
```

## Core concepts

**Model family**: OpenPI implements three model variants from Physical Intelligence:

| Model | Architecture | Speed | Quality | Typical use |
|-------|-------------|-------|---------|-------------|
| pi0 | Flow-matching VLA | Baseline | Highest | Research, complex tasks |
| pi0-fast | Autoregressive action tokens | 2-5x faster | Good | Real-time control |
| pi0.5 | pi0 + improved vision encoder | Baseline | Best | Latest default |

**Key design choices**:
- **Dual backend**: JAX (primary, official training) and PyTorch (community, deployment-friendly)
- **Config-driven**: All training/serving parameters defined in `src/openpi/training/config.py`
- **Norm stats**: Every config requires precomputed normalization statistics before training
- **WebSocket serving**: Policy servers expose a WebSocket API for low-latency inference

**Training loop invariant**: After every config or dataset change, always re-run this cycle:
1. Compute norm stats → 2. Train → 3. Serve checkpoint → 4. Validate inference

## Compute requirements

| Task | GPU | VRAM | Notes |
|------|-----|------|-------|
| Serve pi0.5 (inference) | 1x A100/H100 | ~24 GB | Single GPU sufficient |
| Fine-tune pi0.5 (JAX) | 1x A100 80GB | ~60 GB | Use `fsdp_devices` for multi-GPU |
| Fine-tune pi0 (JAX) | 1x A100 80GB | ~40 GB | Smaller model footprint |
| Fine-tune (PyTorch DDP) | 1-8x A100 | ~40 GB/GPU | torchrun launcher |
| Compute norm stats | CPU or 1x GPU | ~8 GB | Fast, can run on login node |

## Workflow 0: Blank-machine setup

Copy this checklist and track progress:

```text
Setup Progress:
- [ ] Step 1: Clone the public openpi repo with submodules
- [ ] Step 2: Install uv and sync the workspace
- [ ] Step 3: Install the editable package
- [ ] Step 4: Verify core imports and serving entrypoint
```

**Step 1: Clone repo**

```bash
git clone --recurse-submodules https://github.com/Physical-Intelligence/openpi.git
cd openpi
```

If you already cloned without submodules:

```bash
git submodule update --init --recursive
```

**Step 2: Sync dependencies**

```bash
GIT_LFS_SKIP_SMUDGE=1 uv sync
```

**Step 3: Install editable package**

```bash
GIT_LFS_SKIP_SMUDGE=1 uv pip install -e .
```

**Step 4: Verify installation**

```bash
uv run python -c "from openpi.training import config as _config; print(_config.get_config('pi05_droid').name)"
uv run scripts/serve_policy.py --help
```

## When to use vs alternatives

**Use this skill when:**
- Fine-tuning pi0, pi0-fast, or pi0.5 on LeRobot or RLDS datasets
- Serving OpenPI policies for ALOHA, DROID, or LIBERO evaluation
- Converting JAX checkpoints to PyTorch format
- Debugging OpenPI training issues (norm stats, memory, config)

**Use `fine-tuning-openvla-oft` instead when:**
- Fine-tuning OpenVLA with continuous action heads and LoRA
- Reproducing OpenVLA-OFT paper results on LIBERO or ALOHA

**Use `evaluating-cosmos-policy` instead when:**
- Evaluating NVIDIA Cosmos Policy on simulation benchmarks

---

## Workflow 1: JAX fine-tuning on LeRobot data

Copy this checklist and track progress:

```text
JAX Fine-Tuning Progress:
- [ ] Step 1: Select and copy closest training config
- [ ] Step 2: Update dataset mapping and base checkpoint
- [ ] Step 3: Compute normalization statistics
- [ ] Step 4: Launch JAX training
- [ ] Step 5: Serve checkpoint and run inference sanity check
```

**Step 1: Select config**

Copy the closest config from `src/openpi/training/config.py`:

| Config | Use case |
|--------|----------|
| `pi05_libero` | pi0.5 LIBERO fine-tuning |
| `pi0_libero` | pi0 full fine-tuning on LIBERO |
| `pi0_fast_libero` | pi0-fast on LIBERO |
| `pi0_aloha_pen_uncap` | ALOHA custom data |
| `pi05_droid_finetune` | Small custom DROID dataset (LeRobot format) |
| `pi05_full_droid_finetune` | Full DROID RLDS large-scale training |

**Step 2: Update dataset and transforms**

```python
# In src/openpi/training/config.py, modify your config:
TrainConfig(
    name="my_custom_config",
    model_type="pi05",
    data=LeRobotDataConfig(
        repo_id="your-org/your-dataset",
        # Adjust transforms to match your data format
    ),
    weight_loader=Pi05WeightLoader(),  # Match model type
)
```

Set `repo_id` for your dataset and ensure `weight_loader` matches the model type (pi0 vs pi0.5).

**Step 3: Compute normalization statistics**

```bash
uv run scripts/compute_norm_stats.py --config-name <config_name>
```

This must run before every training launch when config, dataset, or transforms change.

**Step 4: Launch JAX training**

```bash
XLA_PYTHON_CLIENT_MEM_FRACTION=0.9 uv run scripts/train.py <config_name> \
  --exp-name=<run_name> \
  --overwrite
```

For full DROID RLDS training, add the `rlds` dependency group:

```bash
uv run --group rlds scripts/compute_norm_stats.py \
  --config-name pi05_full_droid_finetune \
  --max-frames 10000000

XLA_PYTHON_CLIENT_MEM_FRACTION=0.9 uv run --group rlds scripts/train.py \
  pi05_full_droid_finetune \
  --exp-name=<run_name> --overwrite
```

**Step 5: Serve and validate**

```bash
uv run scripts/serve_policy.py policy:checkpoint \
  --p