This model is trained on the Llava-1.5-Instruct dataset. The official
MiniVLA model uses a
single vision encoder (Siglip), which is the key difference.
See the
MiniVLA GitHub README for instructions on how to use this checkpoint for downstream training and finetuning.
1@article{belkhale24minivla,
2 title={MiniVLA: A Better VLA with a Smaller Footprint},
3 author={Suneel Belkhale and Dorsa Sadigh},
4 url={https://github.com/Stanford-ILIAD/openvla-mini}
5 year={2024}
6}
