Deploying Transformers & Building a Semantic Search Engine

Talk 1 Title: Vertex AI Pipelines for training and deploying Huggingface transformers. Speaker: Maximilian Gartz Abstract: This talk touches upon three major developments in contemporary Machine Learning: First, transformers are gaining in popularity, with their applicability extending also to CV tasks. Second, machine learning pipelines that ensure robust and reproducible results become increasingly popular. Every cloud provider has their own toolbox for facilitating the implementation of such pipelines. Lastly, various frameworks try to resolve the problem of how to deploy trained models in a robust and scalable manner. This talk shows how to make use of these developments to implement an easily adaptable end-to-end Vertex AI Pipeline for training Huggingface Transformers models and deploying them with NVIDIAs Triton Inference Server. Bio Maximilian is an ML Engineer at ML6. His main interest lies in MLOps and particularly in end-to-end machine learning pipelines. At ML6 he was leading the first dive into Vertex AI Pipelines right after their release, which led to a wide adoption across projects. He has a strong focus on standardization and thus worked a lot on developing and deploying standard ML pipelines for different ML tasks. Talk 2 Title: Where my docs at? A short story about building a semantic search engine. Speaker: Matthias Richter Abstract: Lexical based information retrieval systems are great for quickly fetching relevant information in a large text corpus. Trad
Abhishek Thakur