changed tokenization

ref_seg_ger.py

@@ -303,8 +303,8 @@ class RefSeg(datasets.GeneratorBasedBuilder):
 
                 yield key, {
                     # "id": f"{os.path.basename(f)}_{chunk_id}",
-                    'input_ids': chunk_ids,
-                    'attention_mask': [1] * len(chunk_ids),
+                    'tokens': chunk_ids,
+                    # 'attention_mask': [1] * len(chunk_ids),
                     # "bbox": split_bboxes,
                     # "RGBs": split_rgbs,
                     # "fonts": split_fonts,