search

dmlc / dgl

Python package built to ease deep learning on graph, on top of existing DL frameworks.
http://dgl.ai
Apache License 2.0
13.43k stars 3k forks source link

Does DGL support exporting models? Similar to pmml/ONNX. #3418

Closed dunesand closed 2 years ago

dunesand commented 2 years ago

In the process of deploying the graph neural network model, it is necessary to export the trained model. Does DGL support this function? If so, what function or package is it?

VoVAllen commented 2 years ago

Currently we don't support them, the main reason is that we have lots of our customized operator that not available in those format. Would you like to deploy dgl inference model in your pipeline, or for other purposes?

dunesand commented 2 years ago

Due to work needs, the GCN model needs to be exported and online on the web.

Currently we don't support them, the main reason is that we have lots of our customized operator that not available in those format. Would you like to deploy dgl inference model in your pipeline, or for other purposes?

Due to work needs, the GCN model needs to be exported and online on the web.

VoVAllen commented 2 years ago

May I ask what kind of work needs it is?

dunesand commented 2 years ago

It is based on GNN's social network anti-fraud work，this requires deploying the trained model on the decision engine.

May I ask what kind of work needs it is?

It is based on GNN's social network anti-fraud work，this requires deploying the trained model on the decision engine.

VoVAllen commented 2 years ago

You can consider directly expose the model by using flask with pure Python for now. Also if the graph and features are fixed, you can consider offline compute them first and then serve it in your cases

Homura2333 commented 2 years ago

I meet a similar problem recently, may I ask which solution did you finally take? Thank you!