Method load

opencompass/datasets/piqa.py:82–112

(path)

    @staticmethod
    def load(path):
        path = get_data_path(path)
        if environ.get('DATASET_SOURCE') == 'ModelScope':
            from modelscope import MsDataset
            dataset = DatasetDict()
            for split in ['train', 'validation']:
                ms_dataset = MsDataset.load(path, split=split)
                dataset_list = []
                for item in ms_dataset:
                    label = item['label']
                    dataset_list.append({
                        'goal':
                        item['goal'],
                        'sol1':
                        item['sol1'],
                        'sol2':
                        item['sol2'],
                        'answer':
                        'NULL' if label < 0 else 'AB'[label]
                    })
                dataset[split] = Dataset.from_list(dataset_list)
        else:
            train_dataset = PIQADatasetV2.load_single(path, 'train.jsonl',
                                                      'train-labels.lst')
            val_dataset = PIQADatasetV2.load_single(path, 'dev.jsonl',
                                                    'dev-labels.lst')
            dataset = DatasetDict({
                'train': train_dataset,
                'validation': val_dataset
            })
        return dataset


@LOAD_DATASET.register_module()