Flexibility wrt tf and keras versions

[pfistfl]

Support multiple versions of keras and tensorflow

[hrvg]

Thanks for this package; it looks extremely promising.

I was wondering if you had any update concerning the support of other tensorflow and keras versions. I am unfamiliar with what could be a major roadblock, but I hit a significant hurdle in my workflow as my system cannot seamlessly install CUDA 10.1 which is required by tensorflow-2.3.0. I have working installs of keras and tensorflow which default to 2.5 and CUDA 11.4.

This would be a significant improvement considering the current defaults from keras and tensorflow and how tedious it is to resort to docker containers or building from source.