support downloading dataset from OpenMind

README.md

@@ -164,6 +164,23 @@ Then submit the evaluation task without downloading all the data to your local d
 humaneval, triviaqa, commonsenseqa, tydiqa, strategyqa, cmmlu, lambada, piqa, ceval, math, LCSTS, Xsum, winogrande, openbookqa, AGIEval, gsm8k, nq, race, siqa, mbpp, mmlu, hellaswag, ARC, BBH, xstory_cloze, summedits, GAOKAO-BENCH, OCNLI, cmnli
 ```
 
+#### (Optional) Automatic Download with OpenMind
+
+Also you can use the [OpenMind](https://modelers.cn/) to load the datasets on demand.
+
+Installation:
+
+```bash
+pip install openmind
+export DATASET_SOURCE=OpenMind
+```
+
+Then submit the evaluation task without downloading all the data to your local disk. Available datasets include:
+
+```bash
+gsm8k
+```
+
 Some third-party features, like Humaneval and Llama, may require additional steps to work properly, for detailed steps please refer to the [Installation Guide](https://opencompass.readthedocs.io/en/latest/get_started/installation.html).
 
 <p align="right"><a href="#top">🔝Back to top</a></p>

README_zh-CN.md

@@ -158,6 +158,22 @@ export DATASET_SOURCE=ModelScope
 humaneval, triviaqa, commonsenseqa, tydiqa, strategyqa, cmmlu, lambada, piqa, ceval, math, LCSTS, Xsum, winogrande, openbookqa, AGIEval, gsm8k, nq, race, siqa, mbpp, mmlu, hellaswag, ARC, BBH, xstory_cloze, summedits, GAOKAO-BENCH, OCNLI, cmnli
 ```
 
+#### (可选) 使用 OpenMind 自动下载
+
+另外，您还可以使用[OpenMind](https://modelers.cn/)来加载数据集：
+环境准备：
+
+```bash
+pip install openmind
+export DATASET_SOURCE=OpenMind
+```
+
+配置好环境后，无需下载全部数据，直接提交评测任务即可。目前支持的数据集有：
+
+```bash
+gsm8k
+```
+
 有部分第三方功能,如 Humaneval 以及 Llama,可能需要额外步骤才能正常运行，详细步骤请参考[安装指南](https://opencompass.readthedocs.io/zh_CN/latest/get_started/installation.html)。
 
 <p align="right"><a href="#top">🔝返回顶部</a></p>

opencompass/datasets/gsm8k.py

@@ -18,9 +18,18 @@ class GSM8KDataset(BaseDataset):
     @staticmethod
     def load(path):
         path = get_data_path(path)
-        if environ.get('DATASET_SOURCE') == 'ModelScope':
+        dataset_source = environ.get('DATASET_SOURCE', None)
+
+        if dataset_source == 'ModelScope':
             from modelscope import MsDataset
             dataset = MsDataset.load(dataset_name=path)
+        elif dataset_source == 'OpenMind':
+            try:
+                from openmind.integrations.datasets import load_dataset
+            except ImportError as ex:
+                raise ImportError('Package `openmind` is required when setting environment variable '
+                                  '`DATASET_SOURCE` to `OpenMind`.') from ex
+            dataset = load_dataset(path)
         else:
             datasets = {}
             for split in ['train', 'test']:

opencompass/utils/datasets.py

@@ -37,6 +37,14 @@ def get_data_path(dataset_id: str, local_mode: bool = False):
         assert ms_id is not None, \
             f'{dataset_id} is not supported in ModelScope'
         return ms_id
+    if dataset_source == 'OpenMind':
+        try:
+            om_id = DATASETS_MAPPING[dataset_id]['om_id']
+        except KeyError as ex:
+            raise KeyError(f"{dataset_id} is not supported in OpenMind.") from ex
+        assert om_id is not None, \
+            f'{dataset_id} is not supported in OpenMind'
+        return om_id
     elif dataset_source == 'HF':
         # TODO: HuggingFace mode is currently not supported!
         hf_id = DATASETS_MAPPING[dataset_id]['hf_id']

opencompass/utils/datasets_info.py

@@ -125,6 +125,7 @@
     "opencompass/gsm8k": {
         "ms_id": "opencompass/gsm8k",
         "hf_id": "opencompass/gsm8k",
+        "om_id": "OpenCompass/gsm8k",
         "local": "./data/gsm8k/",
     },
     # HellaSwag

tests/dataset/test_om_datasets.py

@@ -0,0 +1,173 @@
+import random
+import sys
+import unittest
+import warnings
+from os import environ
+
+from datasets import Dataset, DatasetDict
+from mmengine.config import read_base
+from tqdm import tqdm
+
+from concurrent.futures import ThreadPoolExecutor, as_completed
+
+warnings.filterwarnings('ignore', category=DeprecationWarning)
+
+
+def reload_datasets():
+    modules_to_remove = [
+        module_name for module_name in sys.modules
+        if module_name.startswith('configs.datasets')
+    ]
+
+    for module_name in modules_to_remove:
+        del sys.modules[module_name]
+
+    with read_base():
+        from configs.datasets.gsm8k.gsm8k_gen import gsm8k_datasets
+
+    return sum((v for k, v in locals().items() if k.endswith('_datasets')), [])
+
+
+def load_datasets_conf(source):
+    environ['DATASET_SOURCE'] = source
+    datasets_conf = reload_datasets()
+    return datasets_conf
+
+
+def load_datasets(source, conf):
+    environ['DATASET_SOURCE'] = source
+    if 'lang' in conf:
+        dataset = conf['type'].load(path=conf['path'], lang=conf['lang'])
+        return dataset
+
+    if 'setting_name' in conf:
+        dataset = conf['type'].load(path=conf['path'],
+                                    name=conf['name'],
+                                    setting_name=conf['setting_name'])
+        return dataset
+
+    if 'name' in conf:
+        dataset = conf['type'].load(path=conf['path'], name=conf['name'])
+        return dataset
+
+    try:
+        dataset = conf['type'].load(path=conf['path'])
+    except Exception as ex:
+        print(ex)
+        dataset = conf['type'].load(**conf)
+
+    return dataset
+
+
+def clean_string(value):
+    """Helper function to clean and normalize string data.
+
+    It strips leading and trailing whitespace and replaces multiple whitespace
+    characters with a single space.
+    """
+    if isinstance(value, str):
+        return ' '.join(value.split())
+    return value
+
+
+class TestingOmDatasets(unittest.TestCase):
+
+    def test_datasets(self):
+        # 加载 OpenMind 和 Local 数据集配置
+        om_datasets_conf = load_datasets_conf('OpenMind')
+        local_datasets_conf = load_datasets_conf('Local')
+
+        # 初始化成功和失败的数据集列表
+        successful_comparisons = []
+        failed_comparisons = []
+
+        def compare_datasets(om_conf, local_conf):
+            openmind_path_name = f"{om_conf.get('path')}/{om_conf.get('name', '')}\t{om_conf.get('lang', '')}"
+            local_path_name = f"{local_conf.get('path')}/{local_conf.get('name', '')}\t{local_conf.get('lang', '')}"
+            # 断言类型一致
+            assert om_conf['type'] == local_conf['type'], "Data types do not match"
+            print(openmind_path_name, local_path_name)
+            try:
+                om_dataset = load_datasets('OpenMind', om_conf)
+                local_dataset = load_datasets('Local', local_conf)
+                _check_data(om_dataset, local_dataset, sample_size=sample_size)
+                return 'success', f'{openmind_path_name} | {local_path_name}'
+            except Exception as exception:
+                print(exception)
+                return 'failure', f'{openmind_path_name} is not the same as {local_path_name}'
+
+        with ThreadPoolExecutor(thread) as executor:
+            futures = {
+                executor.submit(compare_datasets, om_conf, local_conf): (om_conf, local_conf)
+                for om_conf, local_conf in zip(om_datasets_conf, local_datasets_conf)
+            }
+
+            for future in tqdm(as_completed(futures), total=len(futures)):
+                result, message = future.result()
+                if result == 'success':
+                    successful_comparisons.append(message)
+                else:
+                    failed_comparisons.append(message)
+
+        # 输出测试总结
+        total_datasets = len(om_datasets_conf)
+        print(f"All {total_datasets} datasets")
+        print(f"OK {len(successful_comparisons)} datasets")
+        for success in successful_comparisons:
+            print(f"  {success}")
+        print(f"Fail {len(failed_comparisons)} datasets")
+        for failure in failed_comparisons:
+            print(f"  {failure}")
+
+
+def _check_data(om_dataset: Dataset | DatasetDict,
+                oc_dataset: Dataset | DatasetDict,
+                sample_size):
+    assert type(om_dataset) == type(
+        oc_dataset
+    ), f'Dataset type not match: {type(om_dataset)} != {type(oc_dataset)}'
+
+    # match DatasetDict
+    if isinstance(oc_dataset, DatasetDict):
+        assert om_dataset.keys() == oc_dataset.keys(
+        ), f'DatasetDict not match: {om_dataset.keys()} != {oc_dataset.keys()}'
+
+        for key in om_dataset.keys():
+            _check_data(om_dataset[key], oc_dataset[key], sample_size=sample_size)
+
+    elif isinstance(oc_dataset, Dataset):
+        # match by cols
+        assert set(om_dataset.column_names) == set(
+            oc_dataset.column_names
+        ), f'Column names do not match: {om_dataset.column_names} != {oc_dataset.column_names}'
+
+        # Check that the number of rows is the same
+        assert len(om_dataset) == len(
+            oc_dataset
+        ), f'Number of rows do not match: {len(om_dataset)} != {len(oc_dataset)}'
+
+        # Randomly sample indices
+        sample_indices = random.sample(range(len(om_dataset)),
+                                       min(sample_size, len(om_dataset)))
+
+        for i, idx in enumerate(sample_indices):
+            for col in om_dataset.column_names:
+                om_value = clean_string(str(om_dataset[col][idx]))
+                oc_value = clean_string(str(oc_dataset[col][idx]))
+                try:
+                    assert om_value == oc_value, f"Value mismatch in column '{col}', index {idx}: {om_value} != {oc_value}"
+                except AssertionError as e:
+                    print(f"Assertion failed for column '{col}', index {idx}")
+                    print(f"om_data: {om_dataset[idx]}")
+                    print(f'oc_data: {oc_dataset[idx]}')
+                    print(f'om_value: {om_value} ({type(om_value)})')
+                    print(f'oc_value: {oc_value} ({type(oc_value)})')
+                    raise e
+    else:
+        raise ValueError(f'Datasets type not supported {type(om_dataset)}')
+
+
+if __name__ == '__main__':
+    sample_size = 100
+    thread = 1
+    unittest.main()