Restructure opengpt-x tasks

[katrinklug]

Restructering OpenGPT-X tasks to locate them in an own folder. Thus we have a better overview about the tasks and can implement upstream changes easier

[KlaudiaTH]

Reminder: Don't forget to move the tasks on branches that are not merged yet

[KlaudiaTH]

@katrinklug
I can't merge the PR since there are build errors caused by missing dependencies (e.g. importlib-resources etc..) This needs to be fixed first. Also, one of the build checks (flake8) failed because of formatting. It is probably checked for PEP-8 formatting rules and so I still ask you to format the code accordingly (I think it should be enough to use the formatting tool black).

[KlaudiaTH]

@katrinklug #87
(I have implemented a patch for this.)

[KlaudiaTH]

@katrinklug There seems to be inconsistencies with some of the arguments. E.g. you included no_tokenizer_check for "hf" and "gpt2" models but missed to also add it to other model types e.g. "hf-causal" or "hf-causal-experimental" or "gpt3" etc.. I have added a temporary fix in the ...thellmann1/workdir/lm_eval_setup/opengptx/lm-evaluation-harness/lm_eval/evaluator.py (from line 67), but this type of args definitely belong to the model args and need to be handled differently:

    # this is just a temporary fix
    if isinstance(model, str):
        if model_args is None:
            model_args = ""
        additional_args = {
            "batch_size": batch_size,
            "device": device,
        }
        if model in ("hf", "gpt2"):
            additional_args["no_tokenizer_check"] = no_tokenizer_check
        lm = lm_eval.models.get_model(model).create_from_arg_string(
            model_args,
            additional_args,
        )

[KlaudiaTH]

@katrinklug #87
I have implemented a patch for this. Can you please apply the patch?
You'll find it in lm_eval_setup/0001-Pass-trust_remote_code-from-model_args-to-AutoTokeni.patch on the lm_eval_setup branch.

e.g.
git am ~/path/to/evaluation/repo/lm_eval_setup/0001-Pass-trust_remote_code-from-model_args-to-AutoTokeni.patch