[Scale ML] Simo Ryu: Minimally Training Video Foundation Model in the wild, from scratch
Speaker
Simo Ryu
Fal.ai
Host
Scale ML
Scale ML
Speaker: Simo Ryu
Topic: Minimally Training Video Foundation Model in the wild, from scratch
Date: Tuesday, Oct 14
Time: 3 PM (EST)
Zoom: https://mit.zoom.us/j/91697262920 (password: mitmlscale)
Abstract:
In this talk, we share our experience on training a >10B diffusion model from scratch. This includes pre-processing, hardware efficiency, architecture, parameterization, and optimizer choices.
Bio:
Simo Ryu is a research lead at fal.ai.