add v1 cpm support

vllm/v1/engine/detokenizer.py

@@ -245,8 +245,9 @@ class FastIncrementalDetokenizer(BaseIncrementalDetokenizer):
 
 class SlowIncrementalDetokenizer(BaseIncrementalDetokenizer):
 
-    def __init__(self, tokenizer: AnyTokenizer, request: EngineCoreRequest):
+    def __init__(self, tokenizer: AnyTokenizer, request: EngineCoreRequest, mode="auto"):
         super().__init__(request)
+        self.mode = mode
 
         self.tokenizer = tokenizer
         params = request.sampling_params
@@ -283,6 +284,7 @@ class SlowIncrementalDetokenizer(BaseIncrementalDetokenizer):
                 skip_special_tokens=self.skip_special_tokens,
                 spaces_between_special_tokens=self.
                 spaces_between_special_tokens,
+                mode=self.mode,
             ))
 
         self.tokens.extend(new_tokens)