Watson Machine Learning for z/OS version 2.4 is our latest major release that was GA’ed back in Q2/2022. The capabilities in this release focus on helping clients use AI to exploit the z16 Integrated Accelerator for AI (AIU) to make faster predictive decisions at scale, and bringing AI to their most mission-critical transactional workloads. This further strengthens our strategy of training anywhere and inferencing on zSystem where your transactions run, including native z/OS applications. Please check out the following highlights to learn more about this new release.  More details on IBM Documentation: https://www.ibm.com/docs/en/wml-for-zos/2.4.0

Below are some of the key features and enhancements we delivered with WMLz 2.4 release :

• Enhanced performance of ONNX model inferencing: By leveraging IBM deep learning compiler (DLC) and exploiting Telum processor and zAIU of z16, performance measurements indicate a reduction in response time greater than 55% (to low single-digit millisecond) and more than 119% throughput improvement when inferencing ONNX models.

• Introduction of ONNX model inferencing natively on z/OS: This feature is available for both the Liberty scoring server on z/OS and through the LINK command for CICS applications. This eliminated the performance overhead caused by running ONNX model inferencing in zCX in our previous release.

• Micro-batch scoring requests of ONNX models: This feature is also available both in the Liberty scoring server and the CICS-integrated scoring server. The micro-batching capability significantly improved the throughput of ONNX model inferencing.

• Customize z/OS Jobnames for WMLz base services and processes: This feature simplifies the management of the application tasks across our client’s enterprise environment.

• Integration between Watson Machine Learning for z/OS and Watson OpenScale on Cloud Pak for Data: This enables our clients to leverage Watson OpenScale to monitor their AI models on z/OS and ensure the trustworthiness of models with fairness, explainability, and robustness

We continue to look into innovative ways to improve the model development process. Currently, our product has an OCP-based WMLz IDE, which will be deprecated soon. A new alternative and lightweight training interface will be introduced for our clients to easily stand up the training environment to leverage Z training runtime. This complements our strategy of supporting the flexibility of using a training environment of our client’s own choice, but at the same time, clients continue to have the choice to develop models using Z training runtime. This new feature is now available through our WMLz V2.4 Q3/2022 updates. You can try out this lightweight feature to leverage Z training runtime.

Our clients continue to tell us that they love the flexibility that Watson Machine Learning for z/OS, provides them, especially when it comes to using any existing data science frameworks to develop their models, such as IBM Cloud Pak for Data’s data science platform or any other open source frameworks. Once models are trained, importing and deploying the models to a z/OS environment is easy using Watson Machine Learning for z/OS for the consuming applications running on zSystems, which accelerates and unlocks in-transaction scoring in real-time for our clients.

If you have questions on getting started with AI on IBM Z, please reach out to us at aionz@us.ibm.com