Schedule
Login

standardising MLOps in a start-up: an example using Dagster and PyTorch

Good processes and toolings allow companies to adapt and strive with changing business requirements and short development cycles. Using Dagster, a new workflow engine, and PyTorch, we found the perfect match to guarantee reliable deployments with fast-changing models. Join us and find out how!

Abstract

Extending a deep learning model beyond your dataset to make it available to multiple use-cases and stakeholders requires a structured workflow. Finding the proper workflow and tooling depends on your use case, but what happens when your use case is dynamic?

In a start-up, changing business requirements are standard, so change the inputs, outputs, and model types: the pipeline may be outdated by the time you have put a “model in production”.

With this talk, we want to show a pipeline based on Dagster, deployed on Kubernetes, and PyTorch, that adds little overhead when making models available to colleagues and projects with minimal requirements, making it flexible to handle very customized workflows if needed.

Speakers
Andrea Giardini & Lorenzo Riches
Track
PyData
Audience Level
Intermediate
Language
English
Duration
30 minutes
Speaker name:
Andrea Giardini
Speaker name:
Lorenzo Riches
      Powered by Vercel Logo