12-08, 14:00–14:30 (UTC), LLM Track
Learn about the different approaches for training large-scale machine learning models using PyTorch.
We will cover a broad overview of tools for training large-scale models using PyTorch. The spectrum of topics would range from "build-it-yourself" approaches (using low-level primitives) to "ready-to-use" approaches (using high-level APIs and libraries). The focus will be on covering the breadth of the landscape instead of going deep into any one approach.
This talk is for Python programmers who are familiar with machine learning and want to familiarize themselves with approaches to scale machine learning. Familiarity with PyTorch is preferred but not a prerequisite, as the scaling strategies are generally standard across the frameworks.
Attendees will walk out of this talk with an understanding of what scaling approaches are helpful for different scenarios and how to leverage these approaches.
Previous knowledge expected
I am a tech lead at Meta. My research is focusing on developing foundation models for multimodal data. Outside of tech, I am interested in economics and finance.