Challenges of Multimodal Feature Stores in AI Training

Join us at our first in-person conference on June 25 all about AI Quality: https://www.aiqualityconference.com/ Accelerating Multimodal AI // MLOps podcast #241 with Ethan Rosenthal, Member of Technical Staff of Runway. Huge thank you to AWS for sponsoring this episode. AWS - https://aws.amazon.com/ // Abstract We’re still trying to figure out systems and processes for training and serving “regular” machine learning models, and now we have multimodal AI to contend with! These new systems present unique challenges across the spectrum, from data management to efficient inference. I’ll talk about the similarities, differences, and challenges that I’ve seen by moving from tabular machine learning, to large language models, to generative video systems. I’ll also talk about the setups and tools that I have seen work best for supporting and accelerating both the research and productionization process. // Bio Ethan works at Runway building systems for media generation. Ethan's work generally straddles the boundary between research and engineering without falling too hard on either side. Prior to Runway, Ethan spent 4 years at Square. There, he led a small team of AI Engineers training large language models for Conversational AI. Before Square, Ethan freelance consulted and worked at a couple ecommerce startups. Ethan found his way into tech by way of a Physics PhD. // MLOps Jobs board https://mlops.pallet.xyz/jobs // MLOps Swag/Merch https://mlops-community.myshopify.com/ // Related Links Website: https://www.ethanrosenthal.com Ethan's mangum opus: https://www.ethanrosenthal.com/2020/08/25/optimal-peanut-butter-and-banana-sandwiches/ Real-time Model Inference in a Video Streaming Environment // Brannon Dorsey // Coffee Sessions #98: https://youtu.be/TNO6rYwP3yg Feature Stores for Self-Service Machine Learning: https://www.ethanrosenthal.com/2021/02/03/feature-stores-self-service/ Gen-1: The Next Step Forward for Generative AI: https://research.runwayml.com/gen1 Machine Learning: The High Interest Credit Card of Technical Debt by D. Sculley et al.: https://research.google/pubs/machine-learning-the-high-interest-credit-card-of-technical-debt/ --------------- ✌️Connect With Us ✌️ ------------- Join our slack community: https://go.mlops.community/slack Follow us on Twitter: @mlopscommunity Sign up for the next meetup: https://go.mlops.community/register Catch all episodes, blogs, newsletters, and more: https://mlops.community/ Connect with Demetrios on LinkedIn: https://www.linkedin.com/in/dpbrinkm/ Connect with Ethan on LinkedIn: https://bsky.app/profile/ethanrosenthal.com

Timestamps: [00:00] Ethan's preferred coffee [00:11] Takeaways [02:07] Falling into LLMs [03:16] Advanced AI Tech Capabilities [04:40] AI-powered video editing tool [06:56] Transition to AI: Diffusion Models [09:09] Multimodal Feature Store breakdown [15:33] Multimodal Feature Stores Evolution [18:09] Benefits of Multimodal Feature Store [25:09] Centralized Training Data Repository [27:33] Large-scale distributed training [32:37 - 33:39] AWS Ad [33:45] Dealing with researchers on productionizing [43:52] Infrastructure for Researchers and Engineers [47:04] Generative DevOps movement [49:21] Structuring teams [52:06] Multimodal Feature Stores Efficiency [54:02] Wrap up