Use the dataset Hub API to fetch tokenizer files

README.md

@@ -138,11 +138,13 @@ The supplied chat template preserves reasoning on every assistant turn.
 After preparing `messages` and `tools` as above, render and tokenize with:
 
 ~~~python
+from huggingface_hub import snapshot_download
 from transformers import AutoTokenizer
 
-tokenizer = AutoTokenizer.from_pretrained(
-    "vfleaking/capysft", subfolder="tokenizer", token=True
+snapshot = snapshot_download(
+    "vfleaking/capysft", repo_type="dataset", allow_patterns="tokenizer/*", token=True
 )
+tokenizer = AutoTokenizer.from_pretrained(f"{snapshot}/tokenizer")
 text = tokenizer.apply_chat_template(
     messages, tools=tools, tokenize=False, add_generation_prompt=False
 )