fix: add clear error message when mistral-common is missing for AutoTokenizer loading Voxtral

Author: junjunjd

src/transformers/models/auto/tokenization_auto.py

@@ -1131,6 +1131,10 @@ def from_pretrained(
 
         model_type = config_class_to_model_type(type(config).__name__)
         if model_type is not None:
+            if model_type == "voxtral" and not is_mistral_common_available():
+                raise ImportError(
+                    "The Voxtral tokenizer requires the 'mistral-common' package. Use `pip install mistral-common` to install the package."
+                )
             tokenizer_class_py, tokenizer_class_fast = TOKENIZER_MAPPING[type(config)]
 
             if tokenizer_class_fast and (use_fast or tokenizer_class_py is None):

tests/models/voxtral/test_tokenization_voxtral.py

@@ -0,0 +1,11 @@
+import pytest
+
+from transformers import AutoTokenizer
+from transformers.models.auto import tokenization_auto
+from transformers.models.voxtral import VoxtralConfig
+
+def test_voxtral_tokenizer_requires_mistral_common(monkeypatch):
+    monkeypatch.setattr(tokenization_auto, "is_mistral_common_available", lambda: False)
+    monkeypatch.setattr(tokenization_auto, "get_tokenizer_config", lambda *args, **kwargs: {})
+    with pytest.raises(ImportError, match="mistral-common"):
+        AutoTokenizer.from_pretrained("dummy", config=VoxtralConfig())