diff --git a/src/molgen3D/training/pretraining/dataprocessing/dataloader.py b/src/molgen3D/training/pretraining/dataprocessing/dataloader.py
index d96c4a8..77548ad 100644
--- a/src/molgen3D/training/pretraining/dataprocessing/dataloader.py
+++ b/src/molgen3D/training/pretraining/dataprocessing/dataloader.py
@@ -630,7 +630,6 @@ def _ensure_tokenizer_ready(self) -> None:
             tokenizer = AutoTokenizer.from_pretrained(
                 str(self._tokenizer_path),
                 use_fast=True,
-                fix_mistral_regex=True,
             )
             sep_id = ensure_tokenizer_pad_token(tokenizer)
             self._tokenizer = tokenizer
diff --git a/src/molgen3D/training/pretraining/torchtitan_runner.py b/src/molgen3D/training/pretraining/torchtitan_runner.py
index e974867..342dec1 100644
--- a/src/molgen3D/training/pretraining/torchtitan_runner.py
+++ b/src/molgen3D/training/pretraining/torchtitan_runner.py
@@ -207,7 +207,6 @@ def _load_tokenizer_details(tokenizer_path: Path) -> TokenizerDetails:
     tokenizer = AutoTokenizer.from_pretrained(
         str(tokenizer_path),
         use_fast=True,
-        fix_mistral_regex=True,
     )
     ensure_tokenizer_pad_token(tokenizer)
     vocab_size = len(tokenizer)