Abstract

In late 2022, there was an alluring promise that large models can do everything - just prompt a sufficiently large model to solve your AI problem.

But after two years of GenAI experimentation, it's clear that even the largest models still fall short for many use cases on quality, speed, cost, or reliability. Enter small language models (SLMs) - nimble, purpose-built models fine-tuned on first-party data to excel at specific use cases.

In this talk, we'll give a concrete framework for thinking about fine-tuning and when it's necessary. Then, we will show how combining modern open-source fine-tuning libraries and clever infrastructure abstractions by Runhouse has made fine-tuning more accessible than ever before.

Speaker

Josh Lewittes

CTO @ Runhouse

Josh is a co-founder and CTO at Runhouse, where he leads the development of Kubetorch. He was previously the ML platform tech lead at Gloat. His work spanned the range of ML activities, across architecting a software development platform for ML, building APIs for new product features, designing and implementing machine learning pipelines for training and production, and supporting auxiliary tooling like feature stores. He has also worked as a developer on OCR systems at KPMG.