This repository was archived by the owner on Mar 11, 2026. It is now read-only.
Repository navigation
This repository was archived by the owner on Mar 11, 2026. It is now read-only.
Import failed when using tensorflow version built from source #574
Description
System information
- OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 16.04
- TensorFlow version and how it was installed (source or binary): 2.0.0 (build from source)
- TensorFlow-Addons version and how it was installed (source or binary): 0.6.0
- Python version: 3.6.8
- Is GPU used? (yes/no): yes
Describe the bug
When I install tensorflow from source code, addons cannot be imported successfully.
NotFoundError: tensorflow_addons/custom_ops/activations/_activation_ops.so: undefined symbol: _ZN10tensorflow12OpDefBuilder4AttrESsReacted by Jesse Farebrother, Julian Schäfer-Zimmermann, Matthias Springstein, Amirhossein Kazemnejad, Kaiyuan Liu, dgronskij, sbushmanov, Polychronis Charitidis and lypbendlf
Activity
seanpmorgan commentedon Oct 7, 2019
Hi @npuichigo thanks for creating this. I have recently noticed this issue and was going to investigate a little more before posting.
This generally occurs from a mismatch in what compiler version was used between TF and TFA. However, it looks like you built TF from source as well. I'll post any findings in this thread as we work through this.
seanpmorgan commentedon Oct 29, 2019
So I plan on continuing looking into this (we should be able to match tfa and tf if both are built on the same machine), but wanted to provide a little more information.
The undefined symbol error occurs because of a mismatch in compilers that causes different symbol name managling. This general issue should be solved by this RFC that is already underway:
https://github.com/tensorflow/community/blob/1c87177cb21ad9e8e48a883b162ff0bedd6da42a/rfcs/20190814-kernel-and-op-registration.md
seanpmorgan commentedon Jan 31, 2020
Consolidating in #987