batched tokenization

docbank.py

@@ -60,7 +60,7 @@ _URLS = {
 _FEATURES = datasets.Features(
     {
         "id": datasets.Value("string"),
-        "input_ids": datasets.Sequence(datasets.Value("string")),
+        "words": datasets.Sequence(datasets.Value("string")),
         "bbox": datasets.Sequence(datasets.Sequence(datasets.Value("int64"))),
         # "RGBs": datasets.Sequence(datasets.Sequence(datasets.Value("int64"))),
         # "fonts": datasets.Sequence(datasets.Value("string")),
@@ -288,7 +288,7 @@ class Docbank(datasets.GeneratorBasedBuilder):
 
                 yield key, {
                     "id": f"{f_id}_{chunk_id}",
-                    'input_ids': split_tokens,
+                    'words': split_tokens,
                     "bbox": split_bboxes,
                     # "RGBs": split_rgbs,
                     # "fonts": split_fonts,