FS#69794 - Tensorflow (apparently affects all recent versions) incompatible with Numpy >1.20

Attached to Project: Community Packages
Opened by Alex Pyattaev (alex.pyattaev) - Friday, 26 February 2021, 12:30 GMT
Last edited by Sven-Hendrik Haase (Svenstaro) - Sunday, 06 June 2021, 21:15 GMT
Task Type Bug Report
Category Packages
Status Closed
Assigned To Sven-Hendrik Haase (Svenstaro)
Konstantin Gizdov (kgizdov)
Architecture x86_64
Severity Low
Priority Normal
Reported Version
Due in Version Undecided
Due Date Undecided
Percent Complete 100%
Votes 1
Private No

Details

Description:

Tensorflow does not play well with numpy >1.20 due to issues in tensorflow itself, no fix available at the moment. At this point hard to come up with any real fix as numpy version was upgraded a while ago.

Additional info:
* See details in upstream: https://github.com/tensorflow/tensorflow/issues/47263
* Downgrading numpy alone does not fix the problem as then everything just tends to segfault

Steps to reproduce:
Install Tensorflow, Create 2 layers and stack them. Observe cryptic errors=)
More details here: https://github.com/tensorflow/tensorflow/issues/47311
This task depends upon

Closed by  Sven-Hendrik Haase (Svenstaro)
Sunday, 06 June 2021, 21:15 GMT
Reason for closing:  No response
Comment by Konstantin Gizdov (kgizdov) - Saturday, 27 February 2021, 10:52 GMT Comment by Sam Linnfer (lllf) - Friday, 19 March 2021, 22:26 GMT Comment by Sven-Hendrik Haase (Svenstaro) - Wednesday, 19 May 2021, 08:51 GMT
Please check out 2.5.0-2. I integrated a patch for numpy 1.20 and it should now be compatible.

Loading...