@chrismattmann No. The weights are originally coming from the tf-hub module https://www.tensorflow.org/hub/modules/google/universal-sentence-encoder/2, this paper(https://arxiv.org/abs/1803.11175) has specified that they have open sourced it. Tf-hub is a python only library, they provide an user friendly way of working with deep learning models with their own caching mechanism and few other stuff, I have exported the metagraphs, weights of universal sentence encoder/2 tf-hub module using tensorflow python; there's a specific way of doing that; I can share the model exporting scripts in a gist if you need. I have uploaded the model to dropbox for now. Perhaps, I can upload it to a more convenient place. How about uscdatascience?
[ Full content available at: https://github.com/apache/tika/pull/248 ] This message was relayed via gitbox.apache.org for [email protected]
