Description:

In this talk, we will explore the revolutionary potential of WebGPU for enhancing the performance and scalability of Large Language Models (LLMs). WebGPU, the next-generation web graphics API, offers unprecedented access to modern GPU features directly from the browser. This enables the development of more efficient, powerful, and scalable machine learning applications. Attendees will gain insights into:

• The basics of WebGPU and its advantages over existing web graphics technologies.
• How WebGPU can be leveraged to optimize the performance of LLMs.
• Practical demonstrations of deploying LLMs using WebGPU.
• Comparative analysis of WebGPU and traditional GPU acceleration techniques.
• Future prospects and challenges in integrating WebGPU with LLMs.

Join us to discover how WebGPU is poised to transform the landscape of browser-based machine learning, offering new possibilities for developers and researchers alike.

Prerequisites:

• You know about some of these LLMs
• and maybe know how to run inference on atleast one LLM, be it any size, any floating bit level.

Video URL:

https://docs.google.com/presentation/d/1S-LS-Te7tqVCChPVWdPIfVrU7kP9zrCRGlEJN2NJoWE/edit?usp=sharing

Content URLs:

Slides

Speaker Info:

Sayantan Das is an ML Engineer at premai.io . Previously he worked for the Vector Institute and Ingenuity Labs in Canada. Also worked as a research intern at ETH Zurich, Indian Space Research Organisation (ISRO) and Indian Statistical Institute (Kolkata).

Speaker Links:

https://www.ucalyptus.me https://huggingface.co/ucalyptus