feat: use ArrowBasedBuilder

wikipedia.py

@@ -419,7 +419,7 @@ class WikipediaConfig(datasets.BuilderConfig):
 _DATE = "20230601"
 
 
-class Wikipedia(datasets.GeneratorBasedBuilder):
+class Wikipedia(datasets.ArrowBasedBuilder):
     """Wikipedia dataset."""
 
     # Use mirror (your.org) to avoid download caps.
@@ -465,19 +465,12 @@ class Wikipedia(datasets.GeneratorBasedBuilder):
             ),
         ]
 
-    def _generate_examples(self, filepath):
+    def _generate_tables(self, filepath):
         """This function returns the examples in the raw (text) form."""
         logger.info("generating examples from = %s", filepath)
 
-        tbl = pq.ParquetFile(filepath)
-        for group_i in range(tbl.num_row_groups):
-            row_group = tbl.read_row_group(group_i)
+        pf = pq.ParquetFile(filepath)
 
-            for batch in row_group.to_batches():
-                for row in zip(*batch.columns):
-                    yield str(row[0]), {
-                        "id": str(row[0]),
-                        "url": str(row[1]),
-                        "title": str(row[2]),
-                        "text": str(row[3]),
-                    }
+        for group_i in range(pf.num_row_groups):
+            tbl = pf.read_row_group(group_i)
+            yield group_i, tbl