Tensorflow 2.1 dependency

40 views
Skip to first unread message

Villem Kand

unread,
Apr 20, 2020, 8:25:30 AM4/20/20
to Google App Engine
I am trying to start a web service through Flask with the app engine, and I require tensorflow 2.1.0 to load a model I trained for this.

However in the requirements.txt when I enter tensorflow==2.1.0 it throws me an error, because the pip version is too old, and I seemingly cannot install the new pip before tensorflow is searched for. What do I do in this situation?

waela

unread,
Apr 21, 2020, 5:56:46 AM4/21/20
to Google App Engine
Hello,

In GAE Standard you cannot change the pip version, however in GAE Flex you can build your own runtime based on a GCP official image and then add a step before installing requirements to force use a specific or latest pip version using something like "python3 -m pip install --user --upgrade pip==9.0.3"

Generically speaking I recommend that you use GAE Flex for Tensorflow, but keep in mind that you need to double-check the logs for any other issues. There might be more things going wrong but it just stops at the first error.

Also, you can follow this Public Issue that seems to be almost identical to stay updated on any progress[1]


Reply all
Reply to author
Forward
0 new messages