OpenCompass/opencompass/datasets/lmeval.py

18 lines
529 B
Python
Raw Normal View History

from typing import List, Optional
from datasets import Dataset, DatasetDict
from opencompass.datasets import BaseDataset
class LMEvalDataset(BaseDataset):
"""A dataset wrapper around the evaluator inputs, designed for
OpenCompass's internal use."""
@staticmethod
def load(predictions: List, references: Optional[List] = None):
content = {'prediction': predictions}
if references:
content['reference'] = references
return DatasetDict(dict(test=Dataset.from_dict(content)))