Santosh Gupta

@JayYip @ash3n now that the tf 2.0 hackathon is over, maybe we should switch to the pytorch huggingface BERT, which is somehow very lightweight. It runs on the colab Tesla...

> > It runs on the colab Tesla k80 GPU no problem. > > TensorFlow and Pytorch are not that different in terms of GPU memory. K80 should be fine...

It's hard to tell what the issue is without knowing your local machine, but common errors are trying to run it in tensorflow 1.0, or if the machine is windows.

wait, are you trying to execute the .ipnb file? It's a notebook file, you need to convert it

Biobert were the initial weights we used, they were just fine tuned on the q/a data. So we haven't had a chance to use the 1.1 weights.

I'm not sure if we have the processed data anymore, as we needed to make room in our drives. Perhaps @ash3n or @JayYip would have the data.

It looks like you used the train your own models notebook, and now you're trying to save the model you trained? Or is it something else?

@Jay 'model.h5' is the name of the file, what that person is doing trying to use the keras function `model.save('model.h5')` to save the weights into Keras's h.5 format. So the...

Can you post of a colab notebook that reproduces the error?

I'm not too sure. I would look up how to convert Bert in general to other languages, since DocProduct's main modules are Bert.