reduce batch size to deal with pyarrow.lib.ArrowInvalid: offset overflow

script.py

@@ -83,6 +83,7 @@ if __name__ == "__main__":
         str(data_path / 'hoho25k_test_x.py'),
         data_files=data_files,
         trust_remote_code=True,
+        writer_batch_size=100
     )
 
     # if TEST_ENV: