Datasets:

---

juletxara

/

mgsm

like 40

Hi, related to that, I am trying to use this dataset as benchmark with lighteval, I am getting this error. I think can be either the problem of the dataset or of the modifications of huggingface. Anyway, trying to load the spanish version ("mgsm:es") gets this:

Traceback (most recent call last):
File "/recipes/test.py", line 210, in test_model
pipeline = Pipeline(
^^^^^^^^^
File "/envs/default/lib/python3.12/site-packages/lighteval/pipeline.py", line 142, in init
self._init_tasks_and_requests(tasks=tasks)
File "/envs/default/lib/python3.12/site-packages/lighteval/pipeline.py", line 222, in _init_tasks_and_requests
LightevalTask.load_datasets(self.tasks_dict, self.pipeline_parameters.dataset_loading_processes)
File "/envs/default/lib/python3.12/site-packages/lighteval/tasks/lighteval_task.py", line 426, in load_datasets
datasets = [task.download_dataset_worker(task) for task in tasks.values()]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/envs/default/lib/python3.12/site-packages/lighteval/tasks/lighteval_task.py", line 453, in download_dataset_worker
dataset = load_dataset(
^^^^^^^^^^^^^
File "/envs/default/lib/python3.12/site-packages/datasets/load.py", line 1492, in load_dataset
builder_instance = load_dataset_builder(
^^^^^^^^^^^^^^^^^^^^^
File "/envs/default/lib/python3.12/site-packages/datasets/load.py", line 1137, in load_dataset_builder
dataset_module = dataset_module_factory(
^^^^^^^^^^^^^^^^^^^^^^^
File "/envs/default/lib/python3.12/site-packages/datasets/load.py", line 1036, in dataset_module_factory
raise e1 from None
File "/envs/default/lib/python3.12/site-packages/datasets/load.py", line 994, in dataset_module_factory
raise RuntimeError(f"Dataset scripts are no longer supported, but found {filename}")
RuntimeError: Dataset scripts are no longer supported, but found mgsm.py

It seems there's a version of this dataset that works: jbross-ibm-research/mgsm