try using manual dir

SNLI-VE.py

@@ -4,6 +4,7 @@
 import json
 import os
 import datasets
+from pathlib import Path
 
 _CITATION = """\
 @article{xie2019visual,
@@ -59,7 +60,6 @@ _SNLI_VE_SPLITS = {
     "validation": "snli_ve_dev.jsonl",
     "test": "snli_ve_test.jsonl",
 }
-JZ_FOLDER_PATH = f"{os.environ['cnw_ALL_CCFRSCRATCH']}/local_datasets/flickr30k-images.tar.gz"
 
 _FEATURES = datasets.Features(
                 {   
@@ -75,7 +75,7 @@ _FEATURES = datasets.Features(
 class SNLIVE(datasets.GeneratorBasedBuilder):
     """SNLIVE."""
 
-    
+    _LOCAL_IMAGE_FOLDER_NAME = "flickr30k-images"
     DEFAULT_CONFIG_NAME = "Default"
     def _info(self):
         return datasets.DatasetInfo(
@@ -87,6 +87,11 @@ class SNLIVE(datasets.GeneratorBasedBuilder):
         )
 
     def _split_generators(self, dl_manager):
+        images_path = os.path.join(
+            dl_manager.extract(os.path.join(dl_manager.manual_dir, "flickr30k-images.tar.gz")),
+            self._LOCAL_IMAGE_FOLDER_NAME
+        )
+
         urls = {
             "Default": {
                 "train": os.path.join(_SNLI_VE_URL_BASE, _SNLI_VE_SPLITS["train"]),
@@ -95,10 +100,6 @@ class SNLIVE(datasets.GeneratorBasedBuilder):
             },
         }
         snli_ve_annotation_path = dl_manager.download_and_extract(urls)
-        images_path = os.path.join(
-            dl_manager.extract(JZ_FOLDER_PATH),
-            "flickr30k-images"
-        )
 
         return [
             datasets.SplitGenerator(