metadata
library_name: gr00t
tags:
- gr00t
- gr00t-n1.7
- robotics
- vla
base_model: nvidia/GR00T-N1.7-3B
groot1.7_tool_insert
Fine-tuned GR00T N1.7-3B on SO101 data (tool_insert).
| Base model | nvidia/GR00T-N1.7-3B |
| Dataset | villekuosmanen/armnetbench_tool_insert |
| Task | tool_insert |
| Training | Isambard-AI GH200, batch 64, action horizon 16 |
| W&B project | groot1.7_tool_insert |
| W&B run | 5sqmj40n |
Checkpoints
| Step | Loss | Path |
|---|---|---|
| 8,000 | 0.0270 | checkpoints/8000/pretrained_model/ |
| 6,000 | 0.0327 | checkpoints/6000/pretrained_model/ |
Usage
Load from a specific step:
from pathlib import Path
# Use checkpoints/<step>/pretrained_model/ with the GR00T inference stack
ckpt = "pravsels/groot1.7_tool_insert/checkpoints/8000/pretrained_model"