I’ve been working on a Postgres extension for running local LLM inference “inside” Postgres (inference runs in a background worker, while Postgres stores the jobs, inputs, outputs, and receipts)
There are a few reasons why this could be chosen over something like Airflow. They are (but not limited to): modern workflow semantics including both functional and imperative API and the ability to work with dynamically mapped tasks, a stricter / richer notion of state, data passage between tasks as a first class operation, better scheduling semantics (workflows aren’t required to have a schedule, scheduled flows can still be run off schedule, and it’s possible to have multiple runs of the same flow occurring at the exact same time), etc…
Hey All, OP here, contrary to popular belief I did not make this :) I saw it on GitHub when I noticed someone who starred the Prefect repo had the username `programming`, clicked their profile and saw this: https://github.com/programming/aaaaaaaaaaaaaaaaaaaaaaaaaaaaa... and just thought it was hilarious. Knew you all would get a kick out of it!