Stanford-ILIAD/minivla-libero90-prismatic

MiniVLA 1B (Prismatic-Compatible Version)

This checkpoint is in a format that is compatible with the training script from the original Prismatic VLMs project codebase, which the OpenVLA team built on top of to develop the OpenVLA model.

This Prismatic-compatible checkpoint may be useful if you wish to fully fine-tune MiniVLA (all 1 billion parameters) via native PyTorch Fully Sharded Data Parallel (FSDP) using the Prismatic VLMs training script. If you instead wish to do Parameter-Efficient Fine-Tuning via LoRA, you can use the MiniVLA checkpoint linked above, which is compatible with the Hugging Face transformers library. We recommend fine-tuning via LoRA if you do not have sufficient compute to fully fine-tune a 1B-parameter model (e.g., multiple A100/H100 GPUs).

Usage Instructions

See the MiniVLA GitHub README for instructions on how to use this checkpoint for full fine-tuning.

Citation

BibTeX:

@article{belkhale24minivla,
    title={MiniVLA: A Better VLA with a Smaller Footprint},
    author={Suneel Belkhale and Dorsa Sadigh},
    url={https://github.com/Stanford-ILIAD/openvla-mini}
    year={2024}
}