Spacy model to tensorflow model

[ande4485]

Hi,

I created a spacy model with ner and text classification. This model works great but I'm wondering if I can export it to tensorflow Lite to test it on a mobile device. Do you think it's possible ? I search with Thinc, TensorflowWrapper etc ... but I didn't understand if it a solution. I didn't find an example for Tensorflow ( only for pytorch).

Many thanks

[svlandeg]

Hi! The spaCy models you'll have trained are in fact Thinc models, and currently there is no functionality available to export those directly to Tensorflow Lite.

What the TensorflowWrapper is supposed to do, is allow you to specify your model in Tensorflow, than load it - through the wrapper - as Thinc model to be used in spaCy. But that's kind of the reverse of what you're looking for :/

[ande4485]

Thank you @svlandeg , but is there an example with tensorflow ? Because there's just an example with Torch.

[svlandeg]

Examples to use the TensorflowWrapper can be found in our test suite: here and here, further examples in Thinc's docs here, and the API specification of the wrapper is here.