Contribute Media
A thank you to everyone who makes this possible: Read More

Modern, typed Python for (multimodal) ML: From training to deployment (Sponsor: Jina.ai)

Description

Typing is at the center of „modern Python“, and tools (mypy, beartype) and libraries (FastAPI, SQLModel, Pydantic, DocArray) based on it are slowly eating the Python world.

This talks explores the benefits of Python type hints, and shows how they are infiltrating the next big domain: Machine Learning

Target audience: Mainly machine learning practitioners that care about improving their code quality and making use of the ever evolving Python ecosystem. This includes people that focus on model training as well as people that focus on model deployment and serving. The secondary target audience is anyone that likes to know more about Python type hints and how they can be helpful in their code base.

Intended takeaways: The audience should leave the talk with three main learnings: - Why Python type hints are useful - Why they are particularly useful in the ML domain - How they can leverage libraries like DocArray in practice

Preliminary outline: The talk can be seen as two parts:

Part 1: Typing in Python - min 0-5: Introduction - min 5-15: Typing and type hints in Python: Short history, and why is it useful? - min 25-25: Tool landscape: Type checkers (mypy, beartype) and other libraries (Pydantic, FastAPI)

Part 2: Python type hints in ML - min 25-40: Why is typing useful in ML? Tensor shapes, multi-modal data, and more - min 40-60: How to get the most out of typing focused tools for ML: jaxtyping and DocArray - How to organize your data using type hints - How to keep track of your tensor shapes using type hints - How to bridge the gap between training and deployment thanks to typing focused libraries

Details

Improve this page