fix string_to_dict usage for windows (#7598)

Author: lhoestq

src/datasets/arrow_dataset.py

@@ -3079,7 +3079,9 @@ def load_processed_shard_from_cache(shard_kwargs: dict[str, Any]) -> Dataset:
             cache_file_with_suffix_pattern = cache_file_prefix + suffix_template + cache_file_ext
 
             for cache_file in glob.iglob(f"{cache_file_prefix}*{cache_file_ext}"):
-                suffix_variable_map = string_to_dict(cache_file, cache_file_with_suffix_pattern)
+                suffix_variable_map = string_to_dict(
+                    Path(cache_file).as_posix(), Path(cache_file_with_suffix_pattern).as_posix()
+                )
                 if suffix_variable_map is not None:
                     file_num_proc = int(suffix_variable_map["num_proc"])
                     existing_cache_file_map[file_num_proc].append(cache_file)

src/datasets/utils/py_utils.py

@@ -175,6 +175,8 @@ def string_to_dict(string: str, pattern: str) -> Optional[dict[str, str]]:
     Args:
         string (str): input string
         pattern (str): pattern formatted like a python f-string
+            This can be a regex - so in case of un-formatting paths you should use posix paths.
+            Otherwise backslashes for windows paths can cause issues.
 
     Returns:
         Optional[dict[str, str]]: dictionary of variable -> value, retrieved from the input using the pattern, or