2.1 KiB
| title | date | draft | images |
|---|---|---|---|
| Offline Pip Packages | 2020-01-20T23:11:05-05:00 | false |
There are a few reasons I can think of to have offline pip packages:
- A package isn't able to compile on a friend's computer since they don't have the million linear algebra libraries that
numpy/scipyrequire. - You want to archive everything to run a piece of software
- You want to control the packages available to a closed network
Regardless, to my surprise, setting up a repository of python wheels doesn't take many steps.
Setup
First I would recommend that you setup a virtual environment. Either through pyenv or python-virtualenv.
Then, install whatever packages you would like. Let us use tensorflow as an example:
pip install tensorflow
We're going to need the packages pip-chill and pip-tools for the next couple steps
pip install pip-chill pip-tools
After you install all the packages you want to be available, freeze the requirements that aren't dependencies to a text file
pip-chill --no-version > requirements.in
We will then use pip-compile in pip-tools to resolve our dependencies and make our packages as fresh as possible.
pip-compile requirements.in
To sync the current virtual environment with the requirements.txt file that gets produced
pip-sync
Now we have a fully working and resolved environment.
From here, we need to install the wheel package to make the binary wheels.
pip install wheel
Then to create the wheels,
pip wheel --wheel-dir=wheels -r requirements.txt
With this you have a whole repository of wheels under the wheels folder!
Client Side
Now you can get all fancy with your deployment, though I just assumed that the files were mounted in some shared folder.
The client can install all the wheels
pip install /path/to/wheels/*
Or they can just install the packages they want
pip install --no-index -f /path/to/wheels/wheels package_name