lovejuly/LaRA-VLA-bridge

Bridge Robot Learning Model

This model is trained on the Bridge dataset for robotic manipulation tasks.

Model Details

• Architecture: SDPA5 with DIT-B
• Learning Rate: 1E-4 → 1E-5
• Batch Size: 16
• Training Steps: 60K
• Configuration: FINAL_NO_IMGLOSS__1.3lr

Files

• model.pt: Model checkpoint (9.7GB)
• config.json: Model configuration
• config.yaml: Training configuration
• dataset_statistics.json: Dataset statistics
• summary.jsonl: Training summary

Usage

import torch

# Load the model
model = torch.load('model.pt')

Citation

If you use this model, please cite the relevant papers.