How to leverage IBM Z Integrated Accelerator for AI with popular# AI libraries on IBM LinuxONE
It’s a very well-established fact that Artificial Intelligence (AI) is going to influence the way businesses operate and touch end consumers in their day to day lives. AI is impacting various industries including but not limited to Financial Services, Government, Healthcare. Both Deloitte* and Mckinsey* have published studies on how important AI is to various industries and governments. Having a robust ecosystem around AI becomes very critical to businesses and organizations to propel them in the next few decades.
IBM recently announced the launch of IBM z16 focusing on Real-time AI for transaction processing at scale. With many popular open source and proprietary AI and ML libraries out there, how can customers preserve their investments in the tool of their choice and yet leverage the on-chip IBM Integrated Accelerator for AI for IBM z16 for real-time inferencing?
This is possible through an open-source library called onnx-mlir. IBM Research and Engineering teams work closely with the onnx-mlir community and have added seamless integration with the IBM Z Integrated AI accelerator. Customers would need to do the following:
1. Use any popular AI library that exports model to the standard ONNX format.
2. Compile the model using onnx-mlir image provided by IBM (link) into a shared library
3. Import the compiled model into your C/C++/Java/Python application and run on IBM Linux on Z
ONNX (Open Neural Network Exchange) is an open-source format used to represent machine learning models. The use of ONNX helps establish a streamlined path to take a project from inception to production by defining a standard format representing a model. Once a model has an ONNX representation it can be deployed to run on any platform that has an ONNX runtime or model compiler. The model is now portable, with no runtime dependencies on the libraries or framework it was trained on; for example, an ONNX model created and trained in TensorFlow can be served without the TensorFlow runtime. The onnx-mlir image is for deep learning models and compiles the ONNX model to produce an executable that is optimized to the IBM Z z16 Integrated Accelerator for AI without changes to the original model itself.
Check out our no-cost onnx-mlir image on IBM Container Registry and contact us on aionz@us.ibm.com for any questions you might have.
*
Sourceshttps://www.forbes.com/sites/louiscolumbus/2020/10/31/the-state-of-ai-adoption-in-financial-services/?sh=218640e92aac
ECONOMIST INTELLIGENCE UNIT STUDY, THE ROAD AHEAD: ARTIFICIAL INTELLIGENCE AND THE FUTURE OF FINANCIAL SERVICES; 2020
https://www2.deloitte.com/us/en/insights/focus/cognitive-technologies/artificial-intelligence-government.html
https://research.aimultiple.com/ai-government/
# - Based on Github stars