Hi, thanks for your great library!
I have question: is it possible to load trained deeppavlov model using only tensorflow?
I trained bert classification model and now there are 7 files: checkpoint, classes.dict, model.data-00000-of-00001, model.index, model.meta, train_log, valid_log. I want to load this model using tensorflow, because with tf library I have more opportunities to manage memory of CPU and GPU.
Hi, @ostreech1997,
The model pipeline consists not only of a TensorFlow model itself but also pre-processing and maybe post-processing components depending on a configuration. So I’m afraid it’s impossible to load the whole pipeline using only TensorFlow. I assume that you’re using a classification config and it uses fastText or GloVe embeddings outside of TensorFlow.
Hi, @ostreech1997 !
Have you handled with your problem?
I have the same one. And I will be very grateful to you if you share with your solution.
Hey @kos345 Thank you very much for your interest!
I strongly advise you against using unsupported old Tensorflow models.
By now we replaced all unsupported Tensorflow models with Pytorch analogue.
If you are interested in ner_rus_bert please use ner_rus_bert_torch
Please let me know if you need further assistance.
Hi, @Vasily Thank you very much for your answer!
I have an already trained model at ner_rus_bert base with Tensorflow engine. And I have to convert it into onnx format.
I couldn’t find any public tools for converting deeppavlov model into onnx. So I want to get model at Tensorflow format or, if it possible, to convert model to ner_rus_bert_torch and then get torch format model. And then I would convert it into onnx.
Could you advise a way to do this? I will be very grateful to you.