OpenVINO™ Toolkit AI Programming Workshops

Large, GenAI models are taking the world by storm. However, recent developments are showing smaller GenAI is equally performant, with the added benefit of being agile, efficient, and not wasting as many resources as their larger counterparts.

In this talk, we’ll discuss intriguing new dilemmas in the GenAI space: Will Large Language Models (LLMs) fit on small format factor machines? Smaller LMs vs. LLMs? Where is the sweet spot for local inference? We’ll discuss the era of LLM compression, including INT8, INT4, and 1-bit LLM models, and how it can be more effective to work on complex deep learning models and big data processing with GPUs, NPUs, and CPUs.

To solve the challenges of GenAI inference, we’ll demonstrate how GenAI deployment on AI PCs and edge devices can be fast and accurate, with the OpenVINO™ toolkit. We’ll explore exciting new developments you can try today with the latest OpenVINO release, including broader LLM models and Mixture of Experts architecture support, new model compression and memory optimization techniques for LLMs; expanded model deployment on Intel HW; deploying AI models with OpenVINO Model Server, and new GenAI workloads!

Join us for an exciting conversation with our AI Evangelists and Fellow from Intel on the future of AI and roadmaps in GenAI, seeing real live demos on how to run LLMs locally for faster, smarter inference, and learn how to do this on your machine today with OpenVINO.

Dive into the New Era AI PC with its unique AI acceleration capabilities that span across CPU, GPU, and NPU. This session provides a comprehensive overview of leveraging the NPU for AI inference tasks and uses OpenVINO™ to optimize the implementation and deployment of AI applications. Experience live demos showcasing the performance and power efficiency of AI applications on the AI PC. Learn how to use OpenVINO to optimize hardware usage for AI inference, enhancing the efficiency and innovative potential of their AI projects.

• Learn what an AI PC is, and its AI acceleration capabilities across CPU, GPU, and NPU.
• Explore a range of AI use cases, including both conventional and generative AI, and see their practical applications on AI PC.
• Discover how the Neural Processing Unit (NPU) enhances AI application performance and power efficiency, especially for tasks traditionally offloaded to AI accelerators.
• Learn the techniques for optimizing and deploying AI applications on NPU using OpenVINO, ensuring maximum power efficiency and performance.
• Experience live demos showing how to seamlessly transition AI applications to the NPU with OpenVINO.
• Find out how developers can experiment and leverage OpenVINO on AI PC to enhance their AI projects.

Learn how to build computer vision and anomaly detection applications for AI inference solutions. Use the open-source Anomalib library for unsupervised learning with imbalanced datasets to handle rare defects in real time. Improve quality control in manufacturing, healthcare, agriculture, and more. See a robot in action and the latest VAND 2.0 Challenge at CVPR.

• Learn about the fundamentals of the OpenVINO™ toolkit components, OpenVINO™ Runtime, and Neural Network Compression Framework (NNCF) in the anomaly detection field.
• Master the integration of various state-of-the-art anomaly detection algorithms using Anomalib 1.0.
• Train and optimize AI models for deployment on edge devices through techniques like quantization and OpenVINO, ensuring efficient performance in resource-constrained environments.
• Gain hands-on experience in model optimization and deployment on edge devices.
• Discuss the future challenges the industry will face for anomaly detection.

Join us to learn about how Hugging Face and Optimum Intel are shaping the future of generative AI by simplifying and optimizing AI inference through the Transformers library.

With just five lines of code, you can now achieve efficient and flexible generative AI. We’ll be showcasing our fundamental efforts in 2023 in bringing optimized workloads such as stable diffusion, large language, and audio transcription models to the AI developer community with a how-to session.

Join us to learn about how Hugging Face and Optimum Intel are shaping the future of generative AI by simplifying and optimizing AI inference through the Transformers library.

With just five lines of code, you can now achieve efficient and flexible generative AI. We’ll be showcasing our fundamental efforts in 2023 in bringing optimized workloads such as stable diffusion, large language, and audio transcription models to the AI developer community with a how-to session.

• Peek into exciting, work-in-progress items and what’s in store for 2024.
• Get updates on the partnership between the Hugging Face and Optimum Intel teams.
• See demos of the latest feature updates with Hugging Face, OpenVINO™, and Optimum Intel:
  • Generative AI and LLM model support via the Hugging Face Optimum Model Hub
  • Training and fine-tuning optimizations for GenAI models
  • Working code snippets for generating text and images, and transcribing audio with our exciting new models.

Explore the deployment of large language models (LLMs) and the practicality and advantages of client PCs and small form factor machines at the edge—with a focus on implementing a conversational voice agent. We’ll delve into the process of efficiently deploying LLMs on resource-constrained devices by examining techniques such as quantization and optimization through OpenVINO™. Additionally, we’ll show how a user-friendly interface built with Gradio can enhance the agent's accessibility and usability.

The agent could serve as a virtual assistant in any industry, from healthcare to manufacturing. If you’re wondering how this is possible, just ask the agent a question!

• Learn how the Edge AI Reference Kit enables the deployment of advanced LLMs on edge devices, breaking new ground in AI accessibility and innovation.
• Understand how OpenVINO optimizes LLMs for AI PCs, ensuring superior performance while maintaining low power consumption.
• Explore how OpenVINO tackles the common hurdles in AI deployment, such as reducing the model size and computational demands to enable efficient large language model applications even in the most resource-constrained edge environments.
• Gain practical skills in implementing LLMs with OpenVINO and take advantage of Intel's tools to create intelligent applications that are both powerful and efficient.
• See how OpenVINO pushes the boundaries of what's possible with LLMs on edge devices, allowing for smarter, faster, and more responsive AI applications.

This workshop is a deep dive into generative AI on the New Era AI PC. Learn how to leverage AI acceleration capabilities for GenAI across CPU, GPU, and NPU. Designed for developers, this session is a comprehensive overview of using GPU and NPU for image-based generative AI and using OpenVINO™ to optimize the implementation and deployment of AI models such as Stable Diffusion, Latent Consistency Models, and more.

We’ll explore parallels and comparisons between different GenAI algorithms and visual, multimodal capabilities on AI PC. This includes text, audio, and images while identifying lessons learned during the development and optimization process. You’ll see live demonstrations showcasing the performance and power efficiency of AI applications on the AI PC, including multimodal applications such as Kosmos-2 and Llava.

• Learn what an AI PC is, and its AI acceleration capabilities across CPU, GPU, and NPU.
• Explore a range of AI use cases, including both conventional and generative AI, and see their practical applications on AI PC.
• Discover how generative AI applications can be powered by CPU, GPU, and NPU heterogeneously to reach maximum performance and power efficiency.
• Experience live demonstrations showing how to seamlessly transition AI applications to the CPU, GPU, and NPU with OpenVINO.
• Discover how developers can experiment with and leverage OpenVINO on AI PC to enhance their generative AI projects.