PyG-HuggingFace Integration - Adding PyG GNN Models to the HuggingFace Model Hub

[ivaylogb]

🚀 The feature, motivation and pitch

The goal is for PyG to support the ability to add PyG-built GNNs into the HuggingFace Graph Learning Model Hub.

As such, we'd support adding PyG GNN Models to: https://huggingface.co/models?pipeline_tag=graph-ml&sort=downloads

Helpful notes: Add the graph-ml pipeline tag in the README metadata, see the following as an example: https://huggingface.co/clefourrier/graphormer-base-pcqm4mv2/blob/main/README.md?code=true#L5

Useful resources:

• How to add models weights to the hub for hosting (which you would just have to tag as graph-ml): https://huggingface.co/docs/hub/models-uploading
• How to add a model to transformers using the new custom code functionality (to allow people to load it directly from the hub using your code): https://huggingface.co/docs/transformers/custom_models
• How to integrate a model to the transformers library (what HF did for Graphormer) https://huggingface.co/docs/transformers/add_new_model

Alternatives

No response

Additional context

No response

[puririshi98]

https://huggingface.co/riship-nv/RGCN/tree/main this is more suited for people training GNNs for research to share their models as pretrained weights