The Next Million AI Apps - Developing Custom Models for Specialized Tasks

Explore the world of custom AI models in this comprehensive workshop presented by Mark Huang at LLMs in Pod Con. Discover how to develop specialized AI applications by leveraging fine-tuning techniques and data-driven approaches like self-refinement. Learn about the challenges of differentiation in a landscape where foundational models are widely accessible, and gain insights into creating unique AI products capable of solving previously unattainable problems. Delve into topics such as task frameworks, data synthesis pipelines, prompt engineering, memory optimization, and parameter-efficient fine-tuning. Benefit from Mark's extensive experience as a Co-founder and Head of AI at Preemo, as well as his background in machine learning at Splunk and Box.

Intro
Workshop Overview
Models
Open AI
GPT
Why Custom Language Models
Task Framework
Helm
Challenges
Data Synthesis Pipeline
Data Synthesis Blog
Prompt Engineering Notebook
Prompt Generator
Memory Footprint
Notebook
Tokenization
Naive Data Parallelization
Parameter Efficient Fine Tuning
Memory Profile
Memory Footprint Comparison
More Memory Usage Tricks