On 7th of April a day long training workshop will focus on providing training and hands-on experience for participants to use the Atlas platform that is part of the HAICGU cluster at Goethe University of Frankfurt. HAICGU pilot cluster installed at the Goethe University of Frankfurt support HPC and AI workloads across the computing continuum. It consist of 28 ARM based compute nodes, one AI training one AI inference node, 2 IO nodes for high performance storage and high speed interconnect. The platform provides development tools and platforms for both AI beginners and experts, helping industry and application partners build scenario-based AI solutions.
Huawei Atlas AI Computing Platform is powered by Huawei Ascend series AI processors, enabling all-scenario AI infrastructure solutions for device-edge-cloud. The Ascend AI platform comprises AI hardware, CANN Heterogeneous compute architecture, MindSpore AI computing framework, and MindX, the Ascend application enablement architecture. Operator developers can directly use CANN to achieve optimal performance. With MindSpore, developers can quickly develop and train models for different scenarios, which are available on devices, edge, and on the cloud. For application developers, MindX is provided to develop fast Ascend-based AI application.
The workshop focuses on
- System overview and Hardware architecture: Introduce the Atlas system portfolio and will introduce the SoC for the acceleration of AI training and inference. Insights into the Da Vinci hardware architecture and its building blocks.
- Software stack: The software stack used to operate the Atlas system portfolio and gives first insights into the software components. The MindSpore deep learning computing framework to accomplish three goals: easy development, efficient execution, and adaptability to all scenarios.
- Installation , training, inference: Introduces the C++ and Python library frameworks required to execute neural networks on the Atlas systems. He gives details of all necessary library call and their call sequence and execution.
- Demonstrations, examples: First insights into the AI training and inference process of the Atlas machines, using the VGG16 as a convolution neural net. Graph Neural Network as an example to give practical insights into the Python framework for AI training and inference. The session aims to provide additional experience for projects which are completely based on the Python programming language.
Lectures and demonstrations will be delivered by
- Heiko Joerg Schick: Chief Architect – Advanced Computing; HiSilicon Turing
- Salli MOUSTAFA, PhD – Software Solution Architect for Advanced Computing at Huawei Technologies
- Jagyan Prasad Mahapatro – AI expert
- Tao Wu – Solutions Architect at Huawei German Research Center
- Karthee Sivalingam – HPC and AI industry development consultant
Date: 7th April 2021
Location: FIAS building, Ruth-Moufang-Straße 1, 60438 Frankfurt am Main
Full details and the time table can be found at SessionLab.
To register, please visit Eventbright.