---
library_name: gr00t
tags:
- gr00t
- gr00t-n1.7
- robotics
- vla
- bimanual
base_model: nvidia/GR00T-N1.7-3B
---

# groot1.7_transfer_cube_20k

Fine-tuned [GR00T N1.7-3B](https://huggingface.co/nvidia/GR00T-N1.7-3B) for `transfer_cube` on SO101 data.

| | |
|---|---|
| **Base model** | [nvidia/GR00T-N1.7-3B](https://huggingface.co/nvidia/GR00T-N1.7-3B) |
| **Dataset** | [villekuosmanen/armnetbench_transfer_cube](https://huggingface.co/datasets/villekuosmanen/armnetbench_transfer_cube) |
| **Task** | `transfer_cube` |
| **Embodiment** | bimanual SO-101 |
| **Training** | 20k steps, batch 32 (8/GPU × 4 DDP), 4× RTX 5090 on Vast.ai, load_bf16=True |
| **Prior HF repo** | [pravsels/groot1.7_transfer_cube](https://huggingface.co/pravsels/groot1.7_transfer_cube) |
| **W&B project** | [groot1.7_transfer_cube](https://wandb.ai/pravsels/groot1.7_transfer_cube) |
| **W&B run** | [6xs8m2yf](https://wandb.ai/pravsels/groot1.7_transfer_cube/runs/6xs8m2yf) |

## Checkpoints

Final checkpoint (step `20000`) lives at the **repository root** for direct loading.

## Usage

Load with the GR00T inference stack from this repo root (weights + processor configs).