Kevin M Jablonka

It is part of the `BigBench` format but it is not used and adds complexity in multiple parts of the codebase (e.g. methods like `_create_task_map`)

They rely on filenames and we now we have two human subsets, making it a bit confusing. I'd removed those methods. I'll also remove their use in the code for...

we now have a bunch of places where the API documentation in the docstring no longer matches the function signature (e.g. `from_model_object` in `PrompterBuilder`. This is confusing and we should...

https://github.com/kjappelbaum/gptchem/blob/615339fd3520edfd03292690a339554d1f27f171/src/gptchem/data.py#L9

For a next sprint, we should work on the docs and follow the [Diátaxis](https://diataxis.fr/) framework. I think there should, in particular, be some more guidance on - how to implement...

- we should drop `mewling_test` - we should perhaps add some of the new models (I guess it could be cool if we add stuff in batches and perhaps, at...