diff --git a/opennyai/summarizer/others/tokenization.py b/opennyai/summarizer/others/tokenization.py
index 116f1a9..0fd62e2 100644
--- a/opennyai/summarizer/others/tokenization.py
+++ b/opennyai/summarizer/others/tokenization.py
@@ -21,12 +21,19 @@
 import unicodedata
 from io import open
 
-from pytorch_transformers import cached_path
+# --- START OF MODIFICATION ---
+# Replaced 'from pytorch_transformers import cached_path' with the modern equivalent from huggingface_hub
+from huggingface_hub import cached_download, HfFolder
 from wasabi import msg
 
-from opennyai.utils.download import CACHE_DIR
+# from opennyai.utils.download import CACHE_DIR # This import is no longer needed/relevant for Hugging Face caching.
+# Hugging Face Hub manages its own cache directories.
+
+# EXTRACTIVE_SUMMARIZER_CACHE_PATH = os.path.join(CACHE_DIR, 'ExtractiveSummarizer'.lower())
+# The cache path should be managed by huggingface_hub directly or specified by environment variables.
+# We will pass cache_dir explicitly to cached_download.
+# --- END OF MODIFICATION ---
 
-EXTRACTIVE_SUMMARIZER_CACHE_PATH = os.path.join(CACHE_DIR, 'ExtractiveSummarizer'.lower())
 
 PRETRAINED_VOCAB_ARCHIVE_MAP = {
     'bert-base-uncased': "https://s3.amazonaws.com/models.huggingface.co/bert/bert-base-uncased-vocab.txt",
@@ -131,30 +138,43 @@ def convert_ids_to_tokens(self, ids):
         return tokens
 
     @classmethod
-    def from_pretrained(cls, pretrained_model_name_or_path, cache_dir=EXTRACTIVE_SUMMARIZER_CACHE_PATH, *inputs,
+    def from_pretrained(cls, pretrained_model_name_or_path, cache_dir=None, *inputs, # Changed default cache_dir to None
                         **kwargs):
         """
         Instantiate a PreTrainedBertModel from a pre-trained model file.
         Download and cache the pre-trained model file if needed.
         """
         if pretrained_model_name_or_path in PRETRAINED_VOCAB_ARCHIVE_MAP:
-            vocab_file = PRETRAINED_VOCAB_ARCHIVE_MAP[pretrained_model_name_or_path]
-        else:
+            vocab_url = PRETRAINED_VOCAB_ARCHIVE_MAP[pretrained_model_name_or_path]
+            # Use huggingface_hub's default cache logic if cache_dir is not provided
+            local_path = cached_download(url=vocab_url, cache_dir=cache_dir)
+            vocab_file = local_path
+        elif os.path.isfile(pretrained_model_name_or_path):
+            # If it's already a local file path, use it directly
             vocab_file = pretrained_model_name_or_path
-        if os.path.isdir(vocab_file):
-            vocab_file = os.path.join(vocab_file, VOCAB_NAME)
-        # redirect to the cache, if necessary
-        try:
-            resolved_vocab_file = cached_path(vocab_file, cache_dir=cache_dir)
-        except EnvironmentError:
+        else:
+            # Assume it's a model name that needs to be resolved by the hub
+            # This might require a model_id and filename. For this old BERT tokenizer,
+            # it's best to stick to the explicit URL or local file path approach.
             msg.fail(
                 "Model name '{}' was not found in model name list ({}). "
                 "We assumed '{}' was a path or url but couldn't find any file "
                 "associated to this path or url.".format(
                     pretrained_model_name_or_path,
                     ', '.join(PRETRAINED_VOCAB_ARCHIVE_MAP.keys()),
-                    vocab_file))
+                    pretrained_model_name_or_path))
             return None
+        
+        # If it's a directory, assume vocab.txt is inside
+        if os.path.isdir(vocab_file):
+            vocab_file = os.path.join(vocab_file, VOCAB_NAME)
+
+        # Ensure the resolved_vocab_file exists
+        if not os.path.isfile(vocab_file):
+             msg.fail(
+                "Resolved vocabulary file '{}' could not be found.".format(vocab_file))
+             return None
+
         # if resolved_vocab_file == vocab_file:
         #     msg.info("loading vocabulary file {}".format(vocab_file))
         # else:
@@ -166,7 +186,7 @@ def from_pretrained(cls, pretrained_model_name_or_path, cache_dir=EXTRACTIVE_SUM
             max_len = PRETRAINED_VOCAB_POSITIONAL_EMBEDDINGS_SIZE_MAP[pretrained_model_name_or_path]
             kwargs['max_len'] = min(kwargs.get('max_len', int(1e12)), max_len)
         # Instantiate tokenizer.
-        tokenizer = cls(resolved_vocab_file, *inputs, **kwargs)
+        tokenizer = cls(vocab_file, *inputs, **kwargs)
         return tokenizer