hack: reverse data file option

src/zeroband/data.py

@@ -32,6 +32,7 @@ class DataConfig(BaseConfig):
     dataset_ratio: Optional[str] = None
     data_rank: Optional[int] = None
     data_world_size: Optional[int] = None
+    reverse_data_files: bool = False
 
 
 class FakeTokenizedDataset(IterableDataset):
@@ -194,6 +195,7 @@ def _load_datasets(
     data_world_size: Optional[int] = None,
     streaming: bool = True,
     probabilities: Optional[List[float]] = None,
+    reverse_data_files: bool = False,
 ) -> Dataset:
     logger.debug(dataset_names)
     ds_args = []
@@ -202,8 +204,11 @@ def _load_datasets(
         _ds_args = {"path": _ds_name}
         if _ds_config:
             _ds_args["name"] = _ds_config
+        _data_files = _get_datafiles(_ds_name, _ds_config, split)
+        if reverse_data_files:
+            _data_files = _data_files[::-1]
+            _ds_args["data_files"] = _data_files
         if data_rank is not None and data_world_size is not None:
-            _data_files = _get_datafiles(_ds_name, _ds_config, split)
             _ds_args["data_files"] = _data_files[data_rank::data_world_size]
         ds_args.append(_ds_args)
 
@@ -244,6 +249,7 @@ def load_all_datasets(data_config: DataConfig, split: str, max_samples: Optional
         data_world_size=data_config.data_world_size,
         streaming=data_config.streaming,
         probabilities=_get_probabilities(data_config),
+        reverse_data_files=data_config.reverse_data_files,
     )
     if max_samples is not None and data_config.streaming:
         if data_config.max_train_samples is not None: