Update: loading script TODO test

utcd.py

@@ -138,14 +138,15 @@ class Utcd(datasets.GeneratorBasedBuilder):
 
         base_path = dl_manager.download_and_extract('datasets.zip')
         split2paths = {s: [os_join(base_path, f'{dir_nm}_split', dnm, f'{s}.json') for dnm in dnms] for s in splits}
+        # order of dataset file paths will be deterministic for deterministic dataset name ordering
         return [
             datasets.SplitGenerator(name=_split2hf_split[s], gen_kwargs=dict(filepath=split2paths[s])) for s in splits
         ]
 
     def _generate_examples(self, filepath: List[str]):
+        id_ = 0
         for path in filepath:  # each file for one split of one dataset
             dnm = path.split(os.sep)[-2]
-            id_ = 0
             with open(path, encoding='utf-8') as fl:
                 dset = json.load(fl)
                 for txt, labels in dset.items():