Deep Learning Models in Production - our presentation on WiMLDS Poznań 20th Meetup
TL;DR: Watch the meetup’s video on YouTube to find out how we reduced costs and speeded up execution time three times on GPU heavy, production-grade workloads! And why we chose AWS Sagemaker Batch Transform to tackle this challenge.
Data Science & Machine Learning Meetup at Fandom
On 30th May 2023, Fandom proudly hosted the WiMLDS Poznań 20th Meetup and Julia from our ML Engineering Team gave one of two exciting presentations. 💡
Deep Learning Models in Production: ML Engineering Perspective
The speech is a technical deep dive into practicalities when it comes to serving multiple deep learning models on a daily basis. It consists of two main parts:
- We start with introducing the project and our current working environment, and explain why we picked Sagemaker Batch Transform Inference (AWS) to efficiently serve various classification models for Fandom articles.
- Second part of presentation describes a couple of optimisation techniques for time and cost-efficient usage of GPU on a large scale. The combination of all techniques yields an impressive result of triple cost reduction! 🚀
Where to watch? 👀
The recording of the presentation is publicly available on YouTube:
WiMLDS 🤝
WiMLDS stands for Women in Machine Learning and Data Science world-wide organisation which mission is to support and promote women and gender minorities who are practicing, studying or are interested in the fields of machine learning and data science.
You can find out more on official website [1] and follow this link for more information about Poznań Chapter.
Originally published at https://dev.fandom.com.
