GR00T N1.5 LoRA — GR-1 Arms-Only CanSort
LoRA adapter fine-tuned from nvidia/GR00T-N1.5-3B on the GR-1 humanoid arms-only CanSort task using the NVIDIA GR00T-X-Embodiment-Sim dataset.
Training details
| Parameter | Value |
|---|---|
| Base model | nvidia/GR00T-N1.5-3B |
| Embodiment | gr1 |
| Data config | fourier_gr1_arms_only |
| Dataset | gr1_arms_only.CanSort (316,878 samples) |
| Steps | 20,000 |
| Batch size | 16 |
| Learning rate | 1e-4 (cosine decay) |
| LoRA rank | 64 |
| LoRA alpha | 128 |
| LoRA dropout | 0.1 |
| Tune diffusion model | false |
| GPU | A100 40GB |
| Final loss | 0.0174 |
| Training time | ~2 hours |
- Downloads last month
- 15
Model tree for shant0602/groot-n1.5-lora-gr1-cansort
Base model
nvidia/GR00T-N1.5-3B