---
library_name: lerobot
tags:
- robotics
- imitation-learning
---

# Multi-Task DiT (diffusion)

LeRobot policy checkpoints uploaded by `goal_gen/upload_hf_checkpoints.sh`.
Each subfolder contains the deployment-ready `pretrained_model/` payload
(model.safetensors + config.json + pre/postprocessor + train_config.json).

| Subfolder | Train step | Final train loss |
|---|---|---|
| `checkpoint-060000` | 60,000 | 0.008 |
| `checkpoint-080000` | 80,000 | 0.007 |
| `checkpoint-100000` | 100,000 | 0.007 |

## Usage

```python
from huggingface_hub import snapshot_download
ckpt_dir = snapshot_download(
    "JayCao99/dit-diffusion-rm65b-stack-v0.0",
    allow_patterns="checkpoint-100000/*",
)
# Then load with the appropriate LeRobot policy class, e.g.:
# from lerobot.policies.multi_task_dit import MultiTaskDiTPolicy
# policy = MultiTaskDiTPolicy.from_pretrained(f"{ckpt_dir}/checkpoint-100000")
```