TensorFlow model support
Opened this issue · 9 comments
Hello there,
First of all, thank you for releasing this amazing product. As a TensorFlow user, I was wondering if you plan to allow to load seq2seq model trained using TensorFlow. Do you think that it would be feasible to extend this project to support TF models?
Alessandro
Hi @aleSuglia, thank you for your interest. While we currently only offer a model API to OpenNMT-py, the visualization server is agnostic to the underlying model. We are currently not planning to develop a tensorflow version, but if you are interested in using your own tensorflow models, all you need to do is develop against the API (PR's are always welcome :) ).
Please let us know if you have any questions, we are happy to further specify what you'd need to do.
Thank you for the answer. I would definitely be interested in trying to extend this program because it may be really useful for my latest project.
+1 for tensorflow model support. I've been using https://github.com/google/seq2seq/, and am moving to the latest seq2seq framework from Google: https://github.com/tensorflow/tensor2tensor/
+1 for tensorflow model support.
@sebastianGehrmann Any updates on this?
No updates and still no plans. We are welcoming any PR's :)
That’s fine to dev against API, however, where’s the API docs that show you how to use the API?
@sebastianGehrmann What is the API for the visualization server, any documentation?
+1 for the TensorFlow model support. I have been using the TensorFlow variant of OpenNMT (https://github.com/OpenNMT/OpenNMT-tf) for a while now along with docker. It would be a great to have the seq-to-seq visualizations integrated seamlessly into the same repo.