Each day of the Intel® Liftoff Days 2024 hackathon featured targeted workshops that delved into the latest advancements in AI technology, focusing on multi-modal AI and quantization.
Participants had the opportunity to learn directly from Intel’s AI engineers and thought leaders, who provided hands-on guidance on building and optimizing models that combine various data types, such as text, images, and audio.
The workshops were designed to be highly interactive, allowing startups to actively engage, ask questions, and collaborate on challenges, and this is exactly what took place during the second workshop of the Hackathon.
As a result, participants left with enhanced capabilities to develop scalable, high-performance AI applications optimized for the edge.
Let’s explore what took place the second workshop
The workshop on Inference of Multimodal Models by Intel® Liftoff mentor, Rahul Nair, was the second workshop that took place at the Intel® Liftoff Days 2024 hackathon. It was an in-depth session focused on optimizing the performance of multimodal AI models, specifically for inference tasks on Intel CPUs and GPUs.
Rahul’s workshop provided valuable insights into maximizing the efficiency of these models, which handle and process different types of data, within a single framework. The session was tailored for AI developers and data scientists looking to leverage Intel’s hardware capabilities to boost their multimodal AI applications.
He broke down the essential elements of a multimodal model, which typically include a vision encoder, a pre-trained language model (LLM), and an embedding projector. The focus is on aligning these components through joint training to seamlessly integrate image and text embeddings.
A significant portion of the workshop was dedicated to demonstrating the use of Intel extensions for PyTorch (IPEX). Rahul highlighted the ipex.optimize function, which automatically adjusts data formats and optimizes models specifically for Intel hardware, ensuring improved performance during inference.
The workshop underscored the importance of optimizing multimodal models to fully leverage Intel CPUs and GPUs, making these models more efficient and practical for a wide range of applications.
Participants recognized the value of collaboration, support, and mutual learning to push the boundaries of what multimodal models can achieve in real-world scenarios.
Rahul Nair concluded the workshop by reiterating the significance of using Intel's hardware optimization tools to enhance the performance and usability of multimodal models.
The session not only provided valuable technical knowledge but also fostered a collaborative spirit among AI startups, encouraging them to explore innovative solutions together.
As a Software Tools Ecosystem Specialist at Intel, I’ve had the privilege of working on the dynamic GenAI initiative. My focus is on driving engagement with the software developer audience. I'm a proud team member of the Intel® Liftoff for Startups and the Developer Engagement, Relations, and Studio team.