Refactor dataset xnli/cnn_dailymail (#1838)

* refactor: add name error for xnli

* add default language

* use name not version

Author: gongenlei

paddlenlp/datasets/cnn_dailymail.py

@@ -188,7 +188,9 @@ def fix_missing_period(line):
     def _get_data(self, mode):
         """ Check and download Dataset """
         dl_paths = {}
-        version = self.config.get("version", "3.0.0")
+        version = self.name
+        if version is None:
+            version = "3.0.0"
         if version not in ["1.0.0", "2.0.0", "3.0.0"]:
             raise ValueError("Unsupported version: %s" % version)
         dl_paths["version"] = version

paddlenlp/datasets/xnli.py

@@ -31,8 +31,10 @@
 from . import DatasetBuilder
 
 __all__ = ['XNLI']
-ALL_LANGUAGES = ("ar", "bg", "de", "el", "en", "es", "fr", "hi", "ru", "sw",
-                 "th", "tr", "ur", "vi", "zh")
+ALL_LANGUAGES = [
+    "ar", "bg", "de", "el", "en", "es", "fr", "hi", "ru", "sw", "th", "tr",
+    "ur", "vi", "zh"
+]
 
 
 class XNLI(DatasetBuilder):
@@ -92,6 +94,12 @@ def _get_data(self, mode, **kwargs):
     def _read(self, filename, split):
         """Reads data."""
         language = self.name
+        if language is None:
+            language = "all_languages"
+        if language not in ALL_LANGUAGES + ["all_languages"]:
+            raise ValueError(
+                f"Name parameter should be specified. Can be one of {ALL_LANGUAGES + ['all_languages']}. "
+            )
         if language == "all_languages":
             languages = ALL_LANGUAGES
         else: