Intel® recently concluded the second edition of its hands-on workshop exclusively designed for AI & ML developers, data scientists, AI researchers and GPU & HPC programmers. Organised in association with Analytics India Magazine, this three-hour-long virtual hands-on workshop showcased how to accelerate PyTorch applications using Intel® Extension of oneAPI AI Analytics toolkit.

The highlight of this compelling workshop was its comprehensive content and hands-on labs on using Intel® Extension for PyTorch* for AI Training. The session exclusively covered the high computational complexity and algorithmic challenge developers face and provided first-hand exposure to Intel® Extension for PyTorch* (IPEX).

Topics covered in the oneAPI workshop:

  • Introduction to Intel® oneAPI AI Analytics toolkit
  • Establishing Intel Dev Cloud for testing & porting applications
  • Overview of Intel hardware features for AI
  • Introduction & overview of Intel® Extensions Pytorch
  • Hands-on labs on using Intel® Extensions Pytorch for AI training and inference
  • Demo of Pytorch model inference using Intel® OpenVINOTM toolkit

The oneAPI workshop kicked off sharp at 9:30 AM with a welcome note by the Developer Marketing Manager- APJ at Intel — Kavita Aroor, introducing the workshop instructors of the day and explaining the key elements of the session, the rules and guidelines of the contests, and audience polls. 

The welcome note was followed by an introductory session of Intel’s oneAPI Ecosystem by Lakshmi Narasimhan, the Technical Consultant Lead at Intel, to get the audience up to speed. The overview was to give the audience an understanding of this cross-architecture language based on C++ and SYCL standards. He spoke about the powerful libraries designed for the acceleration of domain-specific functions and the frameworks and middleware built using oneAPI.

He further explained the top features and benefits of oneAPI AI Toolkit that include — Deep learning performance for training and inference with Intel optimised DL frameworks, pre-trained models and low-precision tools; drop-in acceleration for machine learning and analytics workflows with compute-intensive Python libraries; and seamless scaling of data pipelines across multi-cores, multi-nodes, to optimise end-to-end solutions with cross-architecture support.

To know more about the workshop, click here.

Following that, an expert panel including Jing Xu, Senior Technical Consulting Engineer, working as an AI specialist within the Intel Software group; and Aditya Sirvaiya, an AI Technical Consulting Engineer in Developer Products support & consulting organisation within Intel Software group provided actionable knowledge on Intel architecture hardware features for deep learning and oneDNN library.

Jing started his session by explaining the Non-Uniform Memory Access, a shared memory architecture that describes the placement of main memory modules with respect to processors in a multiprocessor system. 

He also introduced the audience to Intel’s cross-platform performance library of basic building blocks for DL — oneDNN. He explained how it is optimised for Intel architecture processors and processor graphics and supports the computation of various data types.

While Jing spoke about the hardware features for AI and deep neural network libraries, Aditya provided a hands-on Intel optimisation for PyTorch. He talked about how Intel regularly upstreams most of the CPU optimisations to stock PyTorch, using Vectorize kernel by Intel® AVX2/AVX-512. Additionally, he described the IPEX’s ease of use that imports the “intel_pytorch_extension” Python module. The duo also showcased a few exciting case studies on Speech Synthesis and Optical Character Recognition (OCR), and concluded their hands-on labs with training, inference, and a demo on OpenVINO inference.

To check out the learning materials, click here.

To make it an interactive session, the workshop also hosted a Live Q&A session on Discord, conducted by Intel technical experts. One of the major highlights of the workshop was its exclusive contests that included a lucky draw, #oneAPI workshop contest and a DevMesh Project Contest, where attendees had a chance to win exciting prizes and Amazon gift vouchers worth INR 5000/-.

The #oneAPI Workshop Contest was designed to test the participant’s understanding of the AI analytics toolkit. It was judged based on the submissions of the optimised PyTorch code for both training and inference. On the other hand, the DevMesh Project Contest asked attendees to submit their DevMesh project using Intel® DevCloud & oneAPI toolkit on the Intel® DevMesh portal.

The one API workshop also had interesting polls and quizzes, which kept the attendees engaged throughout the session. Finally, the session was concluded with a shout out to the lucky draw winners who were each taking home Amazon Vouchers.

The platform provided attendees with an opportunity to network with expert oneAPI certified instructors to clear their doubts and start a dialogue. At the end of the workshop, all the attendees had the opportunity to share their feedback on the survey link.

______________________________________

DevMesh Project Contest:

Join the Intel® DevMesh, a community portal for developers and creators to share their work & best practices. || Click here — https://devmesh.intel.com.

  • Submit your projects using Intel® DevCloud & oneAPI toolkit on Intel® DevMesh portal by 30 Aug 2021 & email the project link to kavita.aroor@intel.com.
  • If there are projects that use PyTorch, it could be replaced by Intel’s Extension of PyTorch* (for training and inference), or Intel Distribution of OpenVINO™ (for inference) and other relevant oneAPI components. After replacing the components, the performance can be compared with your stock counterparts, which can be submitted for Intel’s evaluation.
  • Projects submitted must include the usage of Intel DevCloud & oneAPI toolkits.

Top 8 projects will stand a chance to win Amazon Gift Voucher worth INR 5000/- each!

Intel is also providing a fantastic opportunity to learn the DPC++ Learning Paths. Click here to know more.

To share your feedback, click on the survey link here.

The post Highlights From Intel® Hands-On Workshop: Accelerate PyTorch Applications Using Intel oneAPI Toolkit appeared first on Analytics India Magazine.