EN

OpenVLA 7B

OpenVLA 7B

Checkpoint

Open 7B vision-language-action model pretrained on Open X-Embodiment.

Default checkpoint

openvla/openvla-7b

openMIT License

Other releases

1

published checkpoints

Supported robots

No verified robot compatibility recorded.

Benchmark results

No published benchmark results.

Specifications

Architecture components: Autoregressive action tokens
Architecture components: VLM backbone
Backbone: Prismatic VLM with DINOv2, SigLIP and Llama 2
Frameworks: Transformers
Frameworks: Transformers / PyTorch
Input modalities: Language instruction
Input modalities: RGB camera images
Languages: English
Model type: Vision-language-action model
Output action space: Normalized 7-DoF end-effector deltas
Parameter count: 7B parameters
Pipeline task: Image-text-to-text
Pipeline task: Robotics
Training data scale: 970K manipulation episodes
Training data sources: Open X-Embodiment
Training data sources: Open X-Embodiment

Lineage

fine tuned from

Source · openvla

Architecture & I/O

ee pose · continuous

7 DoF

Normalized 7-DoF end-effector deltas

Source · openvla

Deployment

PyTorch / Transformers · NVIDIA GPU

bfloat16 · unknown

Source · OpenVLA Collaboration

Transformers / PyTorch · NVIDIA GPU

bfloat16 · unknown

Source · openvla

Training data

pretrain

Open X-Embodiment

970000 episodes · open

Source · openvla

Resources

paper · OpenVLA paper project page · openvla/openvla-7b model card project page · Octo-Small 1.5 training data source repository · OpenVLA official repository hub · openvla/openvla-7b model card weights · openvla/openvla-7b model card

Submit a source or correction