Install GitHub dependency that uses requirements.txt

[mywristbands]

I'd like to install whisper using poetry. I tried doing this by adding the following line to my pyproject.toml file:

whisper = {git = "https://github.com/openai/whisper.git", rev = "0b5dcfdef7ec04250b76e13f1630e32b0935ce76"}

And then I ran poetry update whisper.

While this does work to install whisper, it doesn't install any of whisper's dependencies listed in its requirements.txt. I could manually download all of these dependencies, but this is really something I would expect my package manager to do. Is it possible to use poetry to automatically download whisper and its dependencies?

Or, would I need to switch to another python package dependency manager (such as pipenv) in order to easily install this package?

[neersighted]

Poetry will use the metadata reported by the package's build-system, in this case setuptools. It appears that whisper is attempting to expose their requirements.txt in their metadata (https://github.com/openai/whisper/blob/main/setup.py#L17-L22); building wheels and investigating what metadata is captured in the METADATA file would be your next step to debug.

[mywristbands]

Thank you for the response! I'm glad it seems you believe this repository should work with poetry.

However, I'm not sure how to debug:

• What do you mean by that I should be "building wheels"? Do you mean that I should build the wheel for whisper by following instructions like this?
• How do I investigate the metadata captured in the METADATA file? I don't know what this METADATA file is or where it is located.

I did find an unofficial distribution of whisper on PyPI called whisper.ai. I tried installing it with poetry, and it did install all the dependencies besides PyTorch. So I just manually installed PyTorch, and it's now working. That said, ideally I would like to use the official distribution.

[mywristbands]

For anyone else reading this tread, ultimately I decided to manage my python dependencies with pip, pyenv, pyenv-virtualenv, and a requirements.txt file. Here's a post of mine describing how I set it up.

The reason I decided against Poetry was I was having a lot of trouble getting different wheels of the PyTorch library installed in my different environments. These environments included:

• macOS installed on M1 chip
• linux docker container running on my macOS installed on M1 chip
• linux docker container running on an AWS G4 instance

Using pip made it easy to install the correct PyTorch wheel for each environment.

[tombohub]

So you never found a solution for this?