Openvla 7b Finetuned Libero 10

🚀 OpenVLA 7B Fine-Tuned on LIBERO-10 (LIBERO-Long)

This model is fine - tuned from the OpenVLA 7B model on the LIBERO - 10 dataset, aiming to solve multimodal image - text - to - text tasks in robotics.

✨ Features

• Fine - tuned from the OpenVLA 7B model via LoRA (r = 32) on the LIBERO - 10 (LIBERO - Long) dataset from the LIBERO simulation benchmark.
• Some modifications were made to the training dataset to enhance the final performance. For details, refer to the OpenVLA paper.

🔧 Technical Details

Below are the hyperparameters used for all LIBERO experiments:

• Hardware: 8 x A100 GPUs with 80GB memory
• Fine - tuned with LoRA: use_lora == True, lora_rank == 32, lora_dropout == 0.0
• Learning rate: 5e - 4
• Batch size: 128 (8 GPUs x 16 samples each)
• Number of training gradient steps: 80K
• Quantization: No quantization at train or test time
• Gradient accumulation: No gradient accumulation (i.e., grad_accumulation_steps == 1)
• Shuffle buffer size: shuffle_buffer_size == 100_000
• Image augmentations: Random crop, color jitter (see training code for details)

💻 Usage Examples

Basic Usage

See the OpenVLA GitHub README for instructions on how to run and evaluate this model in the LIBERO simulator.

📄 License

This project is licensed under the MIT license.

📚 Documentation

Citation

BibTeX:

@article{kim24openvla,
    title={OpenVLA: An Open-Source Vision-Language-Action Model},
    author={{Moo Jin} Kim and Karl Pertsch and Siddharth Karamcheti and Ted Xiao and Ashwin Balakrishna and Suraj Nair and Rafael Rafailov and Ethan Foster and Grace Lam and Pannag Sanketi and Quan Vuong and Thomas Kollar and Benjamin Burchfiel and Russ Tedrake and Dorsa Sadigh and Sergey Levine and Percy Liang and Chelsea Finn},
    journal = {arXiv preprint arXiv:2406.09246},
    year={2024}
}