DataHour: Fine Tuning Generative Models

In this session, the audience will get to know fundamental concepts and approaches to fine tune LLM foundation models of various parameter sizes (GPT 8 billion, GPT 43 Billion, GPT 530 Billion). Along with fine tuning, the concept of p-tuning will also be discussed. Fine-tuning LLMs in the economical manner requires use for model parallelism along with data parallelism hence these concepts will also be discussed. Also building an end to end pipeline with LLMs will be discussed by integrating it with an enterprise database through an informed network and prevent LLM from hallucinating .

Prerequisites: No prerequisites, people of any domain or any work experience with an interest in AI can join.

Amit Kumar works as a Senior Enterprise Architect -Generative AI and Deep Learning at NVIDIA. He helps enterprises from various verticals build and architect end to end generative AI and LLM based solutions. He completed his B.Tech in 2010 from IIT Guwahati and is pursuing higher education from Stanford University. He has previously worked at HP research lab, VMware, EFI and got the opportunity to complete internships at Google as a summer of code students, Indian statistical institute Kolkata, Technical University of Braunschweig Germany. He has 7 patents in the areas of Deep learning, NLP, Classical machine learning, statistical modeling, Industry 4.0 and 3D printing.