HARDCODED path to server since this datasets setup is absolute garbage - use at your own risk

rvl_cdip_easyOCR.py

@@ -195,6 +195,9 @@ class RvlCdipEasyOcr(datasets.GeneratorBasedBuilder):
         with open(labels_filepath, encoding="utf-8") as f:
             data = f.read().splitlines()
 
+        # hardcoded to not get stuck in annoying IO and LFS problems in Hub
+        _OCR_DIR = "/cw/liir_data/NoCsBack/jordy/BDPC"
+
         image_to_OCR = self._get_image_to_OCR(_OCR_DIR, split)
         image_to_class_id = self._get_image_to_class_map(data)