These images are built with various backends - Eigen, oneDNN (Compute Library), oneDNN (ArmPL), oneDNN (OpenBLAS) and oneDNN (reference). TensorFlow-Eigen (2.3.0-eigen) is the recommended image while the other images are experimental and any feedback is welcomed.
Note: The docker images use a free version of Arm Performance Libraries, whose use is subject to the terms and conditions of the applicable End User License Agreement (“EULA”)".
Tag description:
| Tags | TensorFlow version | Description |
|---|---|---|
| 2.3.0-eigen | 2.3.0 | (Recommended) TensorFlow 2.3 is compiled with the default Eigen backend. |
| 2.3.0-onednn-acl | 2.3.0 | (Experimental) TensorFlow 2.3 is compiled with oneDNN optimized with Compute Library. |
| 2.3.0-onednn-armpl | 2.3.0 | (Experimental) TensorFlow 2.3 is compiled with oneDNN optimized with Arm Performance Libraries. |
| 2.3.0-onednn-openblas | 2.3.0 | (Experimental) TensorFlow 2.3 is compiled with oneDNN optimized with OpenBLAS. |
| 2.3.0-onednn | 2.3.0 | (Experimental) TensorFlow 2.3 is compiled with oneDNN reference version. |
Please refer to https://github.com/ARM-software/Tool-Solutions/tree/master/docker/tensorflow-aarch64 for the build recipes, examples on how to use the image and to report any issues. For more information on Compute Library, refer to https://github.com/ARM-software/ComputeLibrary.
Content type
Image
Digest
Size
2.5 GB
Last updated
almost 6 years ago
docker pull crefrod/tensorflow-arm-neoverse-n1:2.3.0-onednn-openblas