CODEWITHSUNDEEP
javascriptpytorchtensorflow.jsonnx
Bart
Bart

Reputation: 1087

Best way to run a trained PyTorch LSTM/GRU model fully in the browser

I'm looking into running a trained PyTorch model (containing LSTM/GRU layers) fully in the browser (no backend) as part of an interactive blog post. I've looked at ONNX.js, and that works great, but not for a model containing a GRU layer. I saw someone comment on the ONNX.js github that Gated RNN's are not supported yet, but that was over half a year ago and I can't find any other information about this.

Other than that, it seems like the best option would be to just rewrite the model in Tensorflow and export to Tensorflow.js.

Is there an easier and more direct solution?

Upvotes: 4

Views: 1373

Answers (1)

Santiago M. Quintero
Santiago M. Quintero

Reputation: 1273

There is this thread, that describes the options but is not receiving a lot of attention.

In summary, as of May 2020, there are only two options:

1) ONNX.js but its development is currently stale.

2) Converting the model to Tensorflow.

Technically there is a third one, that includes no server. And that is to run the model in a Mobile Application.

Upvotes: 1

Related Questions