Patch Mistral Tokenizer (#28146)

Signed-off-by: Julien Denize <julien.denize@mistral.ai>

tests/tokenization/test_mistral_tokenizer.py

@@ -334,20 +334,20 @@ class TestMistralTokenizer:
 
     def test_encode(self, mistral_tokenizer: MistralTokenizer):
         token_ids = (
-            [1, 22177, 4304, 2662, 2]
+            [1, 22177, 4304, 2662]
             if mistral_tokenizer.is_tekken
-            else [1, 23325, 2294, 1686, 2]
+            else [1, 23325, 2294, 1686]
         )
 
-        assert mistral_tokenizer.encode("Hello world !") == token_ids[:-1]
-        assert mistral_tokenizer.encode("Hello world !", max_length=3) == token_ids[:-2]
+        assert mistral_tokenizer.encode("Hello world !") == token_ids
+        assert mistral_tokenizer.encode("Hello world !", max_length=3) == token_ids[:-1]
         assert (
             mistral_tokenizer.encode("Hello world !", truncation=True, max_length=3)
-            == token_ids[:-2]
+            == token_ids[:-1]
         )
         assert (
             mistral_tokenizer.encode("Hello world !", truncation=False, max_length=3)
-            == token_ids[:-1]
+            == token_ids
         )
 
         assert (
@@ -358,7 +358,7 @@ class TestMistralTokenizer:
             mistral_tokenizer.encode(
                 "Hello world !", add_special_tokens=True, max_length=3
             )
-            == token_ids[:-2]
+            == token_ids[:-1]
         )
         assert (
             mistral_tokenizer.encode(
@@ -368,7 +368,7 @@ class TestMistralTokenizer:
         )
         assert (
             mistral_tokenizer.encode("Hello world !", add_special_tokens=False)
-            == token_ids[1:-1]
+            == token_ids[1:]
         )
 
     @pytest.mark.parametrize(
@@ -1088,6 +1088,19 @@ class TestMistralTokenizer:
             == expected_tokens[mistral_tokenizer.is_tekken]
         )
 
+    def test_decode_int(
+        self,
+        mistral_tokenizer: MistralTokenizer,
+    ):
+        ids = 1
+        assert (
+            mistral_tokenizer.decode(
+                ids,
+                skip_special_tokens=False,
+            )
+            == "<s>"
+        )
+
     def test_convert_tokens_to_string(self, mistral_tokenizer: MistralTokenizer):
         tokens = (
             [

vllm/transformers_utils/tokenizers/mistral.py

@@ -165,6 +165,7 @@ def _tekken_token_to_id(tokenizer: "Tekkenizer", t: str | bytes) -> int:
 
 class MistralTokenizer(TokenizerBase):
     def __init__(self, tokenizer: "TransformersMistralTokenizer") -> None:
+        from mistral_common.protocol.instruct.validator import ValidationMode
         from mistral_common.tokens.tokenizers.sentencepiece import (
             SentencePieceTokenizer,
         )
@@ -175,6 +176,14 @@ class MistralTokenizer(TokenizerBase):
         self.instruct = self.mistral.instruct_tokenizer
         self.tokenizer = self.instruct.tokenizer
 
+        mode = self.mistral._chat_completion_request_validator._mode
+        if mode != ValidationMode.test:
+            raise ValueError(
+                "Mistral tokenizer must be in test mode. Make sure to "
+                "set `mode='ValidationMode.test'` when creating the "
+                "Mistral tokenizer."
+            )
+
         _mistral_version_str = str(self.tokenizer.version.value)
         self.version: int = int(_mistral_version_str.split("v")[-1])
 
@@ -205,6 +214,7 @@ class MistralTokenizer(TokenizerBase):
     def from_pretrained(
         cls, path_or_repo_id: str, *, revision: str | None = None
     ) -> "MistralTokenizer":
+        from mistral_common.protocol.instruct.validator import ValidationMode
         from transformers.tokenization_mistral_common import (
             MistralCommonTokenizer as TransformersMistralTokenizer,
         )
@@ -212,7 +222,7 @@ class MistralTokenizer(TokenizerBase):
         str_revision = "main" if revision is None else revision
         return cls(
             TransformersMistralTokenizer.from_pretrained(
-                path_or_repo_id, revision=str_revision
+                path_or_repo_id, revision=str_revision, mode=ValidationMode.test
             )
         )
 
@@ -339,15 +349,9 @@ class MistralTokenizer(TokenizerBase):
         max_length: int | None = None,
         add_special_tokens: bool | None = None,
     ) -> list[int]:
-        if add_special_tokens is not None:
-            return self.transformers_tokenizer.encode(
-                text,
-                truncation=truncation,
-                max_length=max_length,
-                add_special_tokens=add_special_tokens,
+        encoded = self.tokenizer.encode(
+            text, bos=add_special_tokens is not False, eos=False
         )
-        else:
-            encoded = self.tokenizer.encode(text, bos=True, eos=False)
 
         if truncation is not False and max_length is not None:
             return encoded[:max_length]
@@ -383,6 +387,9 @@ class MistralTokenizer(TokenizerBase):
         )
 
     def decode(self, ids: list[int] | int, skip_special_tokens: bool = True) -> str:
+        if isinstance(ids, int):
+            ids = [ids]
+
         return self.transformers_tokenizer.decode(
             ids, skip_special_tokens=skip_special_tokens
         )