Docker build of postgresql-15.3 changing the dimension limit for the cube extension, raising it to 350.
This is needed to be able to work with words embedding or other machine learning related vectors with postgres.
Note: since pg350d was released, there has been some efforts to support machine learning friendly vectors in pgvector, which supports up to 16k dimensions.
You can easily generate a build for your own need in term of dimensions by editing this dockerfile.
The cube extension, which you'll use to perform operations on vectors, has a hard limit of 100 dimensions per vector.
A built image is available on Gitlab's registry. You can pull it this way:
docker pull registry.gitlab.com/oelmekki/postgres-350d:15.3
Patching the hardcoded limit is the recommended way in postgres doc.
I've been using it for a few years in production, and didn't encounter any problem.
Even if you're not familiar with docker, reading the Dockerfile from this repos should be easy enough to teach you how to do it yourself. Spoiler:
- sources are in the
contrib/cube/directory from postgres sources - limit is in
contrib/cube/cubedata.h - make and make install should be ran with the
USE_PGXS=trueenv variable