Dependencies & Tensorflow 2.5

I thought I’d update everything by building a new environment, including TF2.5.

Alas, close but no cigar…

Final errors

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
tensorflow-gpu 2.5.0 requires gast==0.4.0, but you have gast 0.3.3 which is incompatible.
tensorflow-gpu 2.5.0 requires grpcio~=1.34.0, but you have grpcio 1.32.0 which is incompatible.
tensorflow-gpu 2.5.0 requires h5py~=3.1.0, but you have h5py 2.10.0 which is incompatible.
tensorflow-gpu 2.5.0 requires tensorflow-estimator<2.6.0,>=2.5.0rc0, but you have tensorflow-estimator 2.4.0 which is incompatible.

and PL installation downgrades urllib3, numpy, Pillow, grpcio, decorator, colorama, h5py, gast.

I must confess I have never looked into dependency management or how dependencies are determined so I would like to understand that and maybe why these are real dependencies.

Just for example: you require numpy 1.19.2 rather than the latest 1.19.5 and I really don’t see that there could/should be incompatibilities in version changes in the 3rd place, and usually incompatibilities are signalled by major version number changes.

Is there anything that can be done to mitigate these issues so that e.g. TF2.5 can be used?

Hi @JulianSMoore,
We have some fixed dependencies because we have (quite a few times) in the past had the issue that one of the dependencies gets updated (even a minor version upgrade) which causes a crash in the tool.
We have gotten a few requests to upgrade to TF2.5 and python 3.8+ so we will likely do that sometime this summer.
Is there anything specific you are looking for from TF2.5? :slight_smile:

Hi @robertl

Apart from the fact that I just like keeping up with releases so I know I can use the latest features, though they may introduce their own bugs they also progressively resolve issues.

However, in this case I was mostly interested in making use of the latest cudatoolkit & cuDNN & nVidia driver for efficiency and speed gains.

Broken by minor version upgrade? Ouch! Bad dependency. No dinner for you until you get cleaned up!

How very annoying, but I understand. Thanks for the insight!