---
library_name: lerobot
base_model: nvidia/GR00T-N1.7-3B
datasets:
- sreetz-nv/so101-clean-up-vials-into-rack-50_20260628_131121
tags:
- lerobot
- robotics
- imitation-learning
- groot
- so101
- gr00t-n1.7
- albumentations
---

# GR00T N1.7 SO-101 Cleanup Vials, Albumentations, H16, 15k

Fine-tuned GR00T N1.7 policy for the SO-101 cleanup-vials-into-rack task.

## Checkpoint

- Source run: `groot-n17-so101-cleanup-vials-relact-bs128-lr1e4-preset-albu-20k-save5k-20260630-1845`
- Checkpoint step: `015000`
- Base model: `nvidia/GR00T-N1.7-3B`
- Dataset: `sreetz-nv/so101-clean-up-vials-into-rack-50_20260628_131121`
- Embodiment tag: `new_embodiment`
- Batch size: `128`
- Chunk size: `16`
- Action steps: `16`
- Optimizer: AdamW, learning rate `1e-4`, weight decay `1e-5`
- Scheduler: cosine decay with warmup, `500` warmup steps
- Relative actions: enabled, excluding `gripper`
- GR00T preprocessor `use_albumentations`: `true`

## Usage

Load this policy with LeRobot using the uploaded Hub repo ID:

```bash
uv run lerobot-eval --policy.path=awrenn53/groot-n17-so101-cleanup-vials-relact-bs128-lr1e4-albu-h16-015000
```