Meta TRIBE v2
Meta FAIR's tri-modal brain encoder — predicts 70K-voxel neural responses
Meta FAIR's tri-modal brain encoder that predicts 70K-voxel neural responses in the human brain — neuroscience meets AI.
Open foundation models — the base layers everything builds on.
Key features
- Multimodal brain encoding model predicting fMRI responses
- Combines text, audio, video into unified Transformer on the cortical surface
- Pretrained weights on Hugging Face; Colab demo with brain visualizations
- Predictions on fsaverage5 mesh of about 20k vertices
- Licensed CC-BY-NC 4.0
Good for
- In-silico neuroscience research
- Predicting brain responses to natural stimuli
Stack & tags
PythonPyTorchneurosciencefmrimultimodalbrain-encodingopen-weightsfoundation-modeltransformer
Quick start
pip install -e .
model = TribeModel.from_pretrained("facebook/tribev2")From the project README on GitHub.