Question about supported framework

Dear All,

I have some experience with BERT and text analysis, but I’m a beginner at :hugs:. And so, please bear my question which might be a little silly.

Here is the situation, I would like to solve a text comprehension task with a proper model.

However, some model seems not supported in TensorFlow according to the following screenshot.

But it also mentioned that “any model saved as before can be loaded back either in PyTorch or TensorFlow” on the “Quick tour” page.

I am a little confused by those two descriptions above. For say, I would like to choose Bert Generation for my task and I have no experience with PyTorch. Would the model work fine or just have some limited function while I fine-tune it in TensorFlow?

I deeply appreciate your kind assistance.


The documentation should be interpreted as “any model saved as before can be loaded back either in PyTorch or TensorFlow as long as there is an implementation for both frameworks”, so no, you wouldn’t be able to use Bert Generation in TensorFlow at all.

1 Like

Dear @sgugger,

Thank you for your prompt explanation. But I have one more question about the following example, showing in the screenshot underneath.

It says that nlptown model only exists in PyTorch, so, it means that I can only make predictions via pipeline, but fine-tuning for nlptown isn’t available in the TensorFlow environment. Is my understanding correct?

Thank you very much for your time.