Hi @astlaan! Here are some tricks that have helped others:

• https://avilpage.com/2020/02/reduce-python-package-footprint.html
• Exclude files with functions configuration

Thank you, this seems helpful.

The compilation approach is interesting, however, I am running Windows. I suppose that compiling the python libraries specifically for a Vercel deployment environment will require a different configuration, so that it will support it instead of just my local machine.

Is there any guide on how to do this for the Vercel case?

Thank you

• Exclude files with functions configuration

How does this differ from using .vercelignore?

How can this exclude help? As far as I know, when deploying in Vercel, the CLI is not uploading the dependencies packages, and Vercel installs the packages itself. Which would make sense since I am running Windows, and I suppose Vercel uses a Linux environment, with could make some C-extended python packages incompatible?.

What would be useful is to allow to drop some files from the dependencies packages during deployment, right after the python packages are installed in Vercel's Linux requirement.

Using the includeFiles and excludeFiles function configuration options and include/exclude files from designated functions. It could be all of them, a subset, or a single function depending on the configurations you set.

Using .vercelignore lets you set files and directories to ignored when uploading your project to Vercel. Sort of like the way .gitignore lets you set files to specify untracked files.

• Configuring functions with vercel.json
• Exclude Files from Deployments with .vercelignore

Those are the commonly used options to control function size. I also found some older posts with examples that might help you:

• https://github.com/orgs/vercel/discussions/103#discussioncomment-6952635
• https://github.com/orgs/vercel/discussions/4354#discussioncomment-8502579

Deploying prebuilt with Vercel CLI could help if you have a way to build locally with reduced function size. I recall others using this in the past for different issues, but I don't have links to any examples at hand.

1) Local build approach

I am having trouble with the local build approach.

1
(vercel-test) PS C:\Users\user\Documents\Code\project> python --version
2
Python 3.12.0
3
(vercel-test) PS C:\Users\user\Documents\Code\project> vercel build
4
Vercel CLI 37.13.0
5
Error: Unable to find any supported Python versions.
6
Learn More: http://vercel.link/python-version
7
(vercel-test) PS C:\Users\user\Documents\Code\project>

It is saying no supported python versions were found. But as you can see, I used precisely version 3.12, as indicated on the provided link (http://vercel.link/python-version):

I found these two pages:

https://github.com/vercel/vercel/discussions/11128

GitHub

Meaning of `Error: Unable to find any supported Python versions.` in Serverless Functions build. · vercel/vercel · Discussion #11128

Question I'm trying to deploy a python serverless function next to a NextJS app. I have my script within api/solver.py and have Pipfile and Pipfile.lock in the project root, with python version 3.9...

https://github.com/orgs/vercel/discussions/6287#discussioncomment-8932365

I tried locally installing both nodejs v18 and v20, along with setting that version in package.json (as indicated in the pages, despite my backend being fully in python), but none of this seemed to worked (unless I missed something), I just kept getting the "Unable to find any supported Python versions". I wish the error was more expressive.

2) Exclude file method

I already removed every single repository file that's not required for the serverless function.

1. Does the exclude file method work with the python depedencies packages? Because the bulk of the size comes from this. Let's say I put some depedencies (ex plotly, pandas) in requirements.txt, and I hit vercel, such that it uploads and is deployed. Will the excludeFiles allow me do drop the files in the plotly/pandas folders that are automatically created by vercel during deployment? Or just files that already existed in the repository pre-deployment?

2. Is the layout in deployment the same as in .vercel/cache that is created with vercel dev?

@amyegan

any input on this?

Try setting the Node.js version to 18.x in project settings. If you're doing anything custom with the files during the build, it might also help to know more about the Build Output API

This is directly relevant to your situation — Vercel has increased the Python Functions bundle size limit to 500MB unzipped (previously 250MB).

Your stack with Plotly, pandas, and related dependencies was sitting at ~360MB, which was over the old 250MB cap but is now comfortably within the new limit. You should be able to skip the dependency-slimming workarounds and deploy as-is!

Reference: https://vercel.com/changelog/python-vercel-functions-bundle-size-limit-increased-to-500mb

Vercel

Python Vercel Functions bundle size limit increased to 500MB - Vercel

The bundle size limit for Python Vercel Functions has increased from 250MB to 500MB, allowing larger dependencies and assets in a single deployment. Learn more in the docs or deploy FastAPI or Flask to get started.