Intel® and Analytics India Magazine have successfully concluded the oneAPI AI Analytics Toolkit Workshop – a master class on Intel® optimisation techniques for accelerating deep learning workloads on March 25, 2022. The workshop, intended for AI and ML developers, data scientists, AI enthusiasts, AI researchers, GPU and HPC programmers, saw more than 200 techies joining this insightful session.
The workshop took the attendees through the Intel® optimisations calibrated for PyTorch*, installation guide, and performance boost number. They also learned about ease-of-use Python API, vectorisation, parallelism, quantisation, operator fusion, constant folding, etc.
Intel® has been working with Facebook (now Meta) to contribute optimisations to the PyTorch*community and commits to continuously optimising PyTorch* with future advancements of Intel® HWs.
Intel® had a demo on the following topics:
- Intel® Optimization for PyTorch*
- Intel® Extension for PyTorch*
- Intel® Extension for PyTorch*targets optimisations on AVX-512 instruction set.
- Intel® Optimization for PyTorch*released in oneAPI AI Analytics Toolkit.
Highlights
Kavita Aroor, the Developer Marketing Manager – APJ at Intel®, started the workshop with a welcome note, introduced the workshop instructors of the day, and explained the rules and guidelines of the contests and audience polls along with the developer ecosystem program.
It was followed by a session on Intel® oneAPI Ecosystem by Aditya Sirvaiya, an AI Technical Consulting Engineer in the Intel® Software group.
Aditya explained that oneAPI is a cross-architecture language based on C++ and SYCL standards. He spoke about the powerful libraries designed to accelerate domain-specific functions and the set of advanced compilers, libraries and porting, analysis and debugger tools that come with it.
Structure of the oneAPI toolkit
Image: Intel®
Intel® oneAPI base toolkit
Aditya described the Intel® oneAPI base toolkit as a core set of tools and libraries for developing high-performance applications on Intel® CPUs, GPUs and FPGAs. It is used by a broad range of developers across industries. He also elaborated on its benefits and features.
- Data parallel C++ compiler, library and analysis tools
- DPC++ Compatibility tool helps users migrate existing code written in CUDA
- Python distribution includes accelerated scikit-learn, NumPy, SciPy libraries
Image: Intel
Intel® AI analytics toolkit powered by oneAPI
Aditya then introduced the attendees to the Intel® AI analytics toolkit powered by oneAPI. It aims to achieve end-to-end performance for data science and AI workloads and is widely used by data scientists, AI researchers, and machine and deep learning developers. It provides drop-in acceleration for ML and analytics workflows with compute-intensive Python libraries. It also gives seamless scaling of data pipelines across multi-cores and multi-nodes to optimise end-to-end solutions with cross-architecture support (Intel® CPUs, GPUs).
Intel® oneAPI Data Analytics Library (oneDAL)
Further, Aditya took the attendees through the Intel® oneAPI Data Analytics Library (oneDAL), which consists of optimised building blocks for all stages of data analytics on Intel® architecture.
What it does:
- Pre-processing-decompression, filtering, normalisation
- Transformation-aggregation, dimension reduction
- Analysis-summary statistics, clustering, etc.
- Modeling-machine learning (training), parameter estimation simulation
- Validation-hypothesis testing, model errors
- Decision making-forecasting, decision trees, etc
Following that, an expert panel by Jing Xu, Senior Technical Consulting Engineer working as an AI specialist within the Intel® Software group, followed, which introduced the attendees to Intel® Optimization for PyTorch*.
Key features and benefits
He listed out more key benefits of Intel® Optimization for PyTorch*.
- Accelerates end-to-end AI and data science pipelines and achieves drop-in acceleration with optimised Python tools built using oneAPI libraries like oneMKL, oneDNN, oneCCL, oneDAL, etc.
- Achieves high performance for deep learning training and inference with Intel®-optimised versions of TensorFlow and PyTorch*, and low-precision optimisation with support for fp16, int8 and bfloat16.
Image: Intel®
Image: Intel®
During the event, Analytics India Magazine also ran a Lucky Draw, wherein ten lucky participants won an Amazon Voucher worth INR 2000/- each at the end of the workshop.
The winners were selected based on their engagement with Discord throughout the workshop. <https://discord.gg/ycwqTP6>
The Lucky Draw winners are:
- Mandeep Suri
- Yogesh Gorane
- Parikshit Rathode
- Rakeshkumar Tammisetti
- Kowsalya Veerabadran
- Shubham Soni
- Sreenivas Chintada
- Medha Sharma
- Alok Srivastava
- Ujwal Kiran