Problem with package installation

[vacquaviva]

Hello there,

We are trying to run a metalearning-based project on the hub and as a starting point we are trying to reproduce Juan's results from this repo. We are using a CPU, PyTorch notebook.

When we run the notebook, we get an error "No module named 'learn2learn'", but we can't install directly via e.g. "!mamba install learn2learn" (error: Could not solve for environment specs
The following package could not be installed
└─ learn2learn does not exist (perhaps a typo or a missing channel).).

As an alternative, we have tried installing the required packages via terminal, but we also get an error "ERROR: Could not build wheels for learn2learn, which is required to install pyproject.toml-based projects".

Does anyone have suggestions for us?

TIA,

Viviana

[jbusecke]

xpost from slack:

I think the best way forward for a project like this (which might have to change python versions) is to build a custom image and use it on the hub.
The docs are not yet very beginner friendly. But I am planning to hold a workshop (sign up here) about this towards the end of the year. That might not fit in your timeline though. If you need to enable this soon, my recommendation is to create a conda environment.yml file with all dependencies needed. You can test that e.g. on your laptop. The important thing is that this needs to be a completely new environment. Once you have that I can help you with setting up a docker image.

[jbusecke]

I just added a github workflow that builds a docker image upon release into metaflux.