- Please visit ibm-zdnn-plugin.
- Overview
- Downloading the IBM Z Optimized for TensorFlow Container Image
- Model Validation
- Execution on the Integrated Accelerator for AI and on CPU
- Using the Code Samples
- Logging Levels
- Removing the IBM Z Optimized for TensorFlow Container Image
- Frequently Asked Questions
- Working with us
- Licenses
- Disclaimer
Welcome to our Open-Beta program for IBM Z Optimized for TensorFlow.
TensorFlow is an open source machine learning framework. It has a comprehensive set of tools that enable model development, training, and inference. It also features a rich, robust ecosystem.
On IBM z16™ hardware, TensorFlow can leverage new inference acceleration capabilities. Specifically, TensorFlow core 2.7 is now available with support for the IBM Integrated Accelerator for AI that leverages the IBM z Deep Neural Network Library. This enablement will allow TensorFlow core 2.7 to transparently target the accelerator for a number of compute-intensive operations. The Open-Beta container image has been released through the IBM Z and LinuxONE Container Registry. Please visit the section Downloading the IBM Z Optimized for TensorFlow Container Image to get started.
Downloading the IBM Z® Optimized for TensorFlow container image requires credentials for the IBM Z and LinuxONE Container Registry, icr.io.
Documentation on obtaining credentials to icr.io is located here.
Once credentials to icr.io are obtained and have been used to login to the registry, you may pull the IBM Z Optimized for TensorFlow container image with the following code block:
# docker or podman*
docker pull icr.io/ibmz/ibmz-optimized-for-tensorflow:1.02In the docker pull command, the version specified above is 1.02. This is based on the version available in the IBM Z and LinuxONE Container Registry.
In the event you need to obtain a new version of the container image, or you want to view additional information about the container image, you may view that information here.
*Note. This documentation will refer to image/containerization commands in terms of Docker. If you are utilizing Podman, please replace docker with podman when using our example code snippets.
The below list of models, but not limited to, trained on x86 or IBM Z, demonstrate focused optimizations that transparently target the accelerator for a number of compute-intensive operations during inferencing.
- BERT
- DenseNet121
- DenseNet169
- DenseNet201
- InceptionResNetV2
- InceptionV3
- NASNetLarge
- ResNet101
- ResNet101V2
- ResNet152
- ResNet152V2
- ResNet50
- Resnet50V2
- VGG16
- VGG19
- Xception
- YOLOv4
Many of these models can be found here.
IBM Z Optimized for TensorFlow follows IBM's train anywhere and deploy on IBM Z strategy.
-
When using IBM Z Optimized for TensorFlow on an IBM z16™ system, TensorFlow core 2.7 graph execution will transparently target the Integrated Accelerator for AI for a number of compute-intensive operations during inferencing with no changes necessary to TensorFlow models.
-
When using IBM Z Optimized for TensorFlow on an IBM z15™ system, TensorFlow core 2.7 will transparently target the CPU with no changes necessary to TensorFlow models.
To manually enforce which execution path should be followed, you may change the environment variable, ZAIUDEVICES, before execution starts.
-
To transparently target the Integrated Accelerator for AI
export ZAIUDEVICES=1orunset ZAIUDEVICES
-
To only transparently target the CPU
export ZAIUDEVICES=0
Note. If problems are encountered when training a model, please export ZAIUDEVICES=0 then try training your model.
- Documentation for our code samples can be found here.
To obtain logging data about inferencing workloads, please use the information below.
The following table lists different logging levels. Each logged message has an associated logging level that categorizes the message.
IBM Z Optimized for TensorFlow includes DTLOG_LEVEL that produces logging messages regarding support for the IBM Integrated Accelerator for AI. This logging information should be used when working with us.
| Level | Description |
|---|---|
| trace | Everything that is occurring in the application |
| debug | Diagnostic information about the application |
| info | General and useful progress information about the application |
| warning | Events that may produce potential problems in the application |
| error | Events that cause the application to perform abnormally |
| fatal | Severe errors that expect the application to abort |
The default DTLOG_LEVEL is info. To change the DTLOG_LEVEL, use the following example code snippet.
The below code snippet sets DTLOG_LEVEL to trace. To set the DTLOG_LEVEL to any other logging level you'd prefer, replace trace with the level of your choice.
export DTLOG_LEVEL=trace
Note. This is not an all-encompassing list of logging methods with respect to TensorFlow. Please reference the TensorFlow documentation for more information about logging.
In the event where you need to remove the IBM Z Optimized for TensorFlow container image, you may follow the below steps.
- Find the image id for the IBM Z Optimized for TensorFlow image by running
docker images - Delete the IBM Z Optimized for TensorFlow image by running
docker rmi IMAGE_ID
If an in-use error occurs while attempting to delete the container image, use the docker ps -a command to show any running containers. Use the docker stop and docker rm commands to remove the running instances of the container. Then re-run the docker rmi command as illustrated above in step 2.
Note. As a reminder, if you are Podman user, please replace docker with podman in the example code snippets provided above.
-
Q: Where can I get the IBM Z Optimized for TensorFlow container image?
- A: Please visit this link here.
-
Q: Why are there multiple TensorFlow container images in the IBM Z and LinuxONE Container Registry?
-
A: You may have seen multiple TensorFlow container images in IBM Z and LinuxONE Container Registry, namely ibmz/tensorflow and ibmz/ibmz-optimized-for-tensorflow.
-
The "ibmz/tensorflow" container image does not have support for the IBM Integrated Accelerator for AI. The "ibmz/tensorflow" only transparently targets the CPU. It does not have any optimizations referenced in this document.
-
The "ibmz/ibmz-optimized-for-tensorflow" container image includes support for TensorFlow core 2.7 graph execution to transparently target the IBM Integrated Accelerator for AI. The "ibmz/ibmz-optimized-for-tensorflow" container image also allows it's users to transparently target the CPU. This container image contains the optimizations referenced in this document and has been released in the Open-Beta program for IBM Z Optimized for TensorFlow.
-
-
-
Q: Where can I run the IBM Z Optimized for TensorFlow container image?
- A: You may run the IBM Z Optimized for TensorFlow container image on IBM Linux on Z or IBM® z/OS® Container Extensions (IBM zCX).
Want to report a bug, request a feature, or provide us feedback?
Contact us directly at aionz@us.ibm.com and become a member of the AI on IBM Z Community.
- Open-Beta (Early Release) license agreement can be found here
- The registered trademark Linux® is used pursuant to a sublicense from the Linux Foundation, the exclusive licensee of Linus Torvalds, owner of the mark on a worldwide basis.
- TensorFlow, the TensorFlow logo and any related marks are trademarks of Google Inc
- Additional license and notice files are within the
licenses-and-noticesdirectory of this repository. You may access the folder here.
IBM's statements regarding its plans, directions and intent are subject to change or withdrawal without notice at IBM's sole discretion. Information regarding potential future products is intended to outline our general product direction and it should not be relied on in making a purchasing decision. The information mentioned regarding potential future products is not a commitment, promise, or legal obligation to deliver any material, code or functionality. Information about potential future products may not be incorporated into any contract. The development, release, and timing of any future features or functionality described for our products remains at our sole discretion.