ModuleNotFoundError: No module named 'tensorflow.contrib'

[TheCodingRecruiter]

I had ran the program initially and I didn't have tensorflow installed and so got a module error telling me that Tensorflow wasn't installed. Then I pip installed it and now I am getting this error. I googled this error, and have found out that TF 1.14 moved contrib to contrib.trt. Since I just installed TF and have the most recent version, this may be why I am encountering the error, but I don't see it listed in the requirements txt on the repo, so I am unsure. Can you assist? For the record, I am trying to run this within pythonanywhere.

[WAUthethird]

You’re positive that you have Tensorflow 1, not 2 installed? GPT-2-Simple doesn’t support TF 2.

[TheCodingRecruiter]

Nope. That must be what it is. I have TF 2. I'll uninstall and install the right version. What version is the last version? TF 1.13?

[WAUthethird]

It is TF 1.15.

[minimaxir]

Yes, that is a TF 2.0 warning, currently this package doesn't support it.

[mobyte0]

Why not add tensorflow==1.14 to requirements.txt?

[minimaxir]

Because tensorflow has two flavors: CPU (tensorflow) and GPU (tensorflow-gpu) and I do not want to force one or the other.

[arnicas]

I just hit this too, and was coming to say add it to requiremnts :) Oh well!

[minimaxir]

I am open to adding a hard assert if the tf package version >= 2.0.0.

[minimaxir]

The assert is added; let me know if the copy needs to be more clear.

[arlyon]

What would it take to port this to 2.0? I have a project that already uses 2.0

[samcamwilliams]

This tripped me up, too. Apparently tensorflow 1.15 supports both CPU and GPU execution. Is there no way that we can add tensorflow==1.15 to the dockerfile?