Transformer models are the powerful neural networks that have become the standard for delivering advanced performance behind these innovations. But there is a challenge: Training these deep learning models at scale and doing inference on them requires a large amount of computing power. This can make the process time-consuming, complex, and costly.

Today we will talk about all kinds of issues around accessible, production level AI solutions. We also talk about ethical questions around AI usage and why open, democratized AI solutions are important.

Learn more:

Hugging Face Hub

Fast Inference on Large Language Models: BLOOMZ on Habana Gaudi2 Accelerator

Accelerating Stable Diffusion Inference on Intel CPUs

Transformer Performance with Intel & Hugging Face Webinar

Intel Explainable AI Tools

Intel Distribution of OpenVINO Toolkit

Intel AI Analytics Toolkit (AI Kit)

Guests:

As the world moves towards more specialized hardware like GPUs, Google TPUs, Habana Gaudi (HPU), FPGAs, etc, enabling developers to run their code on the most efficient hardware can be challenging. Intel oneAPI and xPUs are at the vanguard of this shift and Digital Cortex is leading the way forward with its comprehensive set of products built on these emerging technologies. By providing unprecedented access to compute across varying architectures and platforms, Digital Cortex is helping developers unlock new levels of performance while simplifying development processes.

We talk to Digital Cortex CEO Charlie Wardell to learn how Digital Cortex is building on top of the oneAPI vision to bring a marketplace of compute kernels to their platform that can run anywhere. Charlie also shares his thoughts on entrepreneurship and what made him make the leap from working in a company to founding a company.

Learn more:

oneAPI

Intel oneAPI toolkits

Guest:

Over the last couple of decades, the world of computer graphics has been growing at an amazing rate.  From animated movies to video games, consumers have reaped the benefits of these technological advances.  One of the latest algorithmic advances in rendering is path guiding.  Learn about path guiding and how you can take advantage of this cool technology using Chaos V-Ray and the Intel Path Guiding Library (OpenPGL)

Learn more:

Intel Open Path Guiding Library (OpenPGL) Github

Paper: Importance driven path tracing using the photon map.

Paper: Practical Path Guiding for Efficient Light-Transport Simulation (a.k.a. Practical Path Guiding)

Paper: Path Guiding in Production

See also:

Guests:

Quantum computing technology is becoming more viable, which means programming models are going to have to be created to harness this potentially world changing technology. To many developers, understanding the possibilities of quantum computing and moving from a binary world to a world of quantum states is challenging.

Tony talks to Anne Matsuura, a physicist and Director of Quantum Applications and Architecture in Intel Labs about what quantum computing can do for society and how Intel is working on building hardware and software solutions to make that future a reality.

We also talk about what a quantum computer actually looks like today (hint: it’s not a tiny flat chip).

Learn more:

oneAPI

Guests:

AI is the new hot thing in business. Everyone wants to talk about how they are harnessing AI to make their business better. However this is not a trivial process, and maybe companies are left wondering what went wrong. Aible is a company devoted to providing business impact in less than 30 days.

We talk to Arijit Sengupta, the CEO of Aible, about how Aible is able to make an impact quickly. He talks about how Aible can transform an AI solution into a business solution, using high schoolers and Aible versus data scientists to prove the point. We also discuss bias in AI, the future of AI, and how AI has the power to transform how we work and live.

Learn more:

Aible/Intel blog

Intel/Aible Performance Benchmark and Case Studies Report

Case Study Videos: How to guarantee impact from AI in 30 days

AI is a Waste of Money

Intel AI Analytics Toolkit

Intel Distribution of OpenVINO™ Toolkit

oneAPI

Guests:

As AI becomes more prevelent in society, one of the biggest questions is how can we take AI and harness it for good? One area that is extremely active is healthcare. AI is being explored across a variety of disciplines in the medical field.

One way to use AI in healthcare is to make doctors more efficient, and that is precisely what Chest-rAi aims to do. Learn a little bit about how AI can be harnessed to make it easier for doctors to be more efficient and help more patients through use of Chest-rAi.

Learn more:

Chest-rAi

Intel AI Analytics Toolkit

Intel Distribution of OpenVINO™ Toolkit

oneAPI

The accelerator landscape has been rapidly changing over the past decade, and as super computers come online, they face the challenge of supporting the latest and greatest accelerators. Through use of C++. Kokkos and SYCL, the engineers working on the Aurora project at Argonne National Laboratory are demonstrating how to enable workloads that are performant now, while continuing to keep an eye on the future.

Guests:

James Reinders – Technical Evangelist @ Intel

Learn more:

Argonne Leadership Computing Facility and Aurora

SYCL Tech

oneAPI

Kokkos

Today, Codeplay Software announces expanding oneAPI for Nvidia and AMD GPUs, making multiarchitecture, multivendor programming much easier. This includes new plugins: oneAPI for Nvidia GPUs and oneAPI for AMD GPUs (beta).

Software developers are increasingly looking to move away from proprietary interfaces and use open standards that are defined by the industry. This can allow them to target a broad set of platforms and architectures. oneAPI aims to bring this to reality by supporting the widest set of processors and accelerators. As a cross-industry initiative, oneAPI offers an open, standards-based, multiarchitecture programming model and specification.

Having official plugins for Nvidia and AMD platforms is a major milestone for the oneAPI open community. The oneAPI Nvidia and AMD GPU plugins seamlessly work with the Intel oneAPI DPC++/C++ Compiler 2023 and many popular libraries, bringing open, standards-based programming to a range of processors from multiple hardware vendors. The oneAPI for Nvidia GPUs and oneAPI for AMD GPUs packages are available to download for free on the Codeplay website.

Learn more:

Codeplay

oneAPI for NVIDIA GPUs

oneAPI for AMD GPUs (beta)

James’ blog on the Intel oneAPI 2023 Toolkit release

Codeplay blog on the oneAPI for NVIDIA and AMD plugins

ChatGPT, DALL-E, Stable Diffusion and other generative AI have rapidly captivated the public. They provide new, easy ways, to create new content based on AI models that have been trained on large, public data sets. How does this technology work? What does it mean for society? In this podcast we discuss the technology behind generative AI, some of the potential legal and societal issues, and what the pros and cons of generative AI are.

For more information see:

Attention is All You Need

Copilot

ChatGPT

Stable Diffusion

Midjourney

openVINO

Intel Extension for PyTorch

oneAPI

Tim Davis @ Texas A&M Github Copilot Tweet

ChatGPT Tech Radar Article

Guests

Parallel programming models are often used on large scale systems, with huge datasets and a large amount of obvious parallelism. In this podcast, Tony talks to two university students that are building a chat bot using consumer hardware and standard frameworks like TensorFlow and oneAPI. They use TensorFlow to enable their AI model and SYCL to optimize various parts of their data pipeline.

They also talk about the Intel Student Ambassador program, Intel Discord, and how Josh and Harvey became part of the Intel social community.

For more information see:

TensorFlow

SYCL

oneAPI

Building LLVM to compile SYCL for NVIDIA GPUs

Intel Student Ambassador Program

Intel Discord

OpenAI – GPT-2, GPT-3 resources

Google LaMDA

Kaggle

AI Predicts Heart Disease Risk Using Single X-Ray

Guests