Push_to_hub() makes Inference Endpoint not working

There’s no progress bar or any indication when using the push_to_hub callback in keras/tensorflow- making it kind of a pain.

To mitigate this, I used model.push_to_hub() and tokenizer.push_to_hub() after training.

But for some reason, after uploading to the hub, the API gives me an Internal Server Error. Yet somehow, using the Trainer API does not give me this problem. Furthermore, the Trainer API correctly determines the task (translation) and the former classifies it as a Text2TextGeneration (using the same model checkpoint).

So how do I properly upload my model to the hub using tensorflow/keras? I’m not sure where I am doing wrong here but for more information, I was just following this notebook translation-tf on my own colab file.

This seems to be unrelated to inference endpoint, Would open a thread in the #beginners section?