By Prashantha Subbarao, Andrew Sica, Pradipta Ghosh 

In November of last year we announced the AI Toolkit for IBM Z and LinuxONE, which provides IBM Elite support for a key optimized set of open-source AI libraries and model servers. This offering is designed to enable enterprises to leverage these frameworks in production with confidence. Information about the offerings supported within the AI Toolkit for IBM Z and LinuxONE can be found here. 

We are pleased to announce the addition of IBM Z Accelerated for PyTorch to the list of AI frameworks supported by the AI Toolkit. IBM Z Accelerated for PyTorch features the ability to exploit IBM Z Integrated Accelerator for AI on IBM z16 and LinuxONE 4. This provides improved inference performance for a wide range of PyTorch models. You can read more about accelerating PyTorch inference on IBM Z and IBM LinuxONE in our blog.

We have continued our commitment to the existing offerings by both delivering new capabilities and ensuring open-source packages are kept up to date. In particular, these new capabilities include features designed to further optimize real-time inference processing. This includes:

• Optimized data preprocessing: Enhancements in IBM Snap ML enable you to export scikit-learn data preprocessing pipelines for use in Triton inference server with the IBM Snap ML C++ backend. This avoids the overhead of python based preprocessing, enabling lower latency and improved scalability.
•       Improved graph optimizations in TensorFlow, TF-Serving and Deep Learning Compiler for a wide range of AI models.

Key challenges addressed by AI Toolkit for IBM Z and LinuxONE

Open source has emerged as a powerful force in driving innovation and accessibility across various fields. It has had a significant impact on the world of artificial intelligence (AI), playing a key role in its evolution.

Many of the enterprises relying on IBM Z and LinuxONE are embracing open-source software to modernize application development and leverage new technologies such as AI. They have grown significant organizational skills in open-source AI frameworks that can be leveraged across their hybrid environment. However, unsupported open source can lead to security and compliance risk; this is especially concerning for sensitive production environments. The AI Toolkit for IBM Z and LinuxONE is a perfect solution that provides a supported, high-performing, and validated environment to help mitigate challenges around open-source support for the enterprise AI solutions.

Benefits of AI Toolkit for IBM Z and LinuxONE

AI Toolkit for IBM Z and LinuxONE includes highly optimized runtimes and serving environments that can leverage all the software and hardware capabilities provided within the IBM Z platform. These are capabilities that are focused on bringing best of the breed solutions in a form factor that can be used readily and easily. They are instrumented to transparently and seamlessly leverage the IBM Z Integrated Accelerator for AI on IBM z16 and LinuxONE 4 for significantly faster inference. In addition to that, AI Toolkit for IBM Z and LinuxONE provides enterprise grade elite service and helps bring solutions based on open source into production on an enterprise system. Apart from service and support, we have put these open-source AI libraries through a secure engineering process, scanned and vetted them and constantly monitor them for vulnerabilities. Coupling all these together, AI Toolkit for IBM Z and LinuxONE is not only a capability that is high performant but also trustworthy.

Components of AI Toolkit for IBM Z and LinuxONE

• IBM Z Accelerated for TensorFlow - Popular Machine Learning and Deep Learning (ML/DL) lifecycle management platform optimized to run on IBM Z and leverage IBM Z Integrated Accelerator for AI. 
• IBM Z Accelerated Serving for TensorFlow - Flexible and high-performing serving platform for ML/DL models optimized to leverage IBM Z Integrated Accelerator for AI.
• IBM Z Accelerated for SnapML - A library that optimizes the training/scoring of popular ML models optimized to leverage IBM Z Integrated Accelerator for AI.
• IBM Z Deep Learning Compiler - Generates a program from any Open Neural Network Exchange (ONNX) DL models to execute on z/OS or Linux on Z optimized to leverage IBM Z Integrated Accelerator for AI.
• IBM Z Accelerated for NVIDIA Triton Inference Server - High-performance inference server that supports the deployment of ML or DL models at scale optimized to leverage IBM Z Integrated Accelerator for AI.
• IBM Z Accelerated for PyTorch – A popular Machine Learning framework based on Torch library, used for applications such as language processing and computer vision and optimized to run on IBM Z and leverage IBM Z Integrated Accelerator for AI.

How to get started

IBM Z and LinuxONE Container Image Registry (ICR) includes open-source software in container images that are often used as the foundation for new composite workloads. It provides a secure and trustworthy content source. On the IBM Z and LinuxONE Container Registry, these components of AI Toolkit for IBM Z and LinuxONE are freely available. Information regarding technical support can be found here.

IBM offers a no-charge discovery workshop that is designed to enable clients on AI on IBM Z and LinuxONE technologies and provide them with project planning and implementation guidance. Additionally, IBM has created AI Solution Templates to accelerate your time to value by providing easy to use full solution samples.  AI Solution Templates are a suite of pre-built blueprints that guide you through the full AI lifecycle on IBM Z with various enterprise use cases while leveraging a variety of technologies.