learned-sparse-retrieval
learned-sparse-retrieval copied to clipboard
Enable pip install?
I'm wondering are you planning to add a setup.py to allow pip install?
I could draft one if you want :)
This would be nice! I'm happy to help test/draft it.
I'm not too familiar with Hydra, but is there a way to reference a yaml file outside of the package directory? If we want to make this a package, not requiring people to modify the content of the package when extending to other models/experiments would be a better workflow.
I think you can use --config-dir <dir> to add an additional directory to the search path.
would it be more convenient to provide method to train without yaml file config?
I would say maybe... depends on what use cases you imagine people would be adapting this framework to.
The current setup makes it very easy to replicate experiments and adapt the code to new data, but more complicated to modify the code or inherit the classes. The latter is making this framework more like a toolkit (e.g. huggingface transformers). I certainly like to use it like a toolkit but it really depends on where you imagine this repository is heading to.