[Model] Add optional parameter to reasoning parser constructor (#25554)

Signed-off-by: taohui <taohui3@gmail.com>
Signed-off-by: Tao Hui <taohui3@gmail.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

vllm/reasoning/abs_reasoning_parsers.py

@@ -34,7 +34,7 @@ class ReasoningParser:
     It is used to extract reasoning content from the model output.
     """
 
-    def __init__(self, tokenizer: AnyTokenizer):
+    def __init__(self, tokenizer: AnyTokenizer, *args, **kwargs):
         self.model_tokenizer = tokenizer
 
     @cached_property

vllm/reasoning/basic_parsers.py

@@ -35,8 +35,8 @@ class BaseThinkingReasoningParser(ReasoningParser):
         """The token that ends reasoning content."""
         raise NotImplementedError
 
-    def __init__(self, tokenizer: AnyTokenizer):
+    def __init__(self, tokenizer: AnyTokenizer, *args, **kwargs):
-        super().__init__(tokenizer)
+        super().__init__(tokenizer, *args, **kwargs)
 
         if not self.model_tokenizer:
             raise ValueError(

vllm/reasoning/glm4_moe_reasoning_parser.py

@@ -26,8 +26,8 @@ class Glm4MoeModelReasoningParser(ReasoningParser):
     from the model's output.
     """
 
-    def __init__(self, tokenizer: PreTrainedTokenizerBase):
+    def __init__(self, tokenizer: PreTrainedTokenizerBase, *args, **kwargs):
-        super().__init__(tokenizer)
+        super().__init__(tokenizer, *args, **kwargs)
         self.think_start_token = "<think>"
         self.think_end_token = "</think>"
 

vllm/reasoning/gptoss_reasoning_parser.py

@@ -24,8 +24,8 @@ class GptOssReasoningParser(ReasoningParser):
     is only used for detecting the end of the reasoning content.
     """
 
-    def __init__(self, tokenizer: PreTrainedTokenizerBase):
+    def __init__(self, tokenizer: PreTrainedTokenizerBase, *args, **kwargs):
-        super().__init__(tokenizer)
+        super().__init__(tokenizer, *args, **kwargs)
         self.reasoning_end_token_ids = self.model_tokenizer.encode(
             "<|start|>assistant<|channel|>final<|message|>")
 

vllm/reasoning/granite_reasoning_parser.py

@@ -24,8 +24,8 @@ class GraniteReasoningParser(ReasoningParser):
     and "Here is my response:" to separate its thinking / response outputs.
     """
 
-    def __init__(self, tokenizer: PreTrainedTokenizerBase):
+    def __init__(self, tokenizer: PreTrainedTokenizerBase, *args, **kwargs):
-        super().__init__(tokenizer)
+        super().__init__(tokenizer, *args, **kwargs)
 
         # NOTE: There have been some observed occurrences of quantized
         # instances of the current models using "Here's" instead of "Here is",

vllm/reasoning/hunyuan_a13b_reasoning_parser.py

@@ -40,8 +40,8 @@ class HunyuanA13BReasoningParser(ReasoningParser):
     response ends: "\n</answer>": [524, 9399, 29]
     """
 
-    def __init__(self, tokenizer: PreTrainedTokenizerBase):
+    def __init__(self, tokenizer: PreTrainedTokenizerBase, *args, **kwargs):
-        super().__init__(tokenizer)
+        super().__init__(tokenizer, *args, **kwargs)
         self.think_start_expr = r"<think>\n"
         self.think_end_expr = r"\n</think>\n"
 

vllm/reasoning/mistral_reasoning_parser.py

@@ -21,12 +21,12 @@ class MistralReasoningParser(DeepSeekR1ReasoningParser):
     text. This parser extracts the reasoning content from the model output.
     """
 
-    def __init__(self, tokenizer: MistralTokenizer):
+    def __init__(self, tokenizer: MistralTokenizer, *args, **kwargs):
         if not isinstance(tokenizer, MistralTokenizer):
             raise ValueError(
                 "The tokenizer must be an instance of MistralTokenizer.")
 
-        ReasoningParser.__init__(self, tokenizer)
+        ReasoningParser.__init__(self, tokenizer, *args, **kwargs)
 
         if not self.model_tokenizer:
             raise ValueError(

vllm/reasoning/step3_reasoning_parser.py

@@ -24,8 +24,8 @@ class Step3ReasoningParser(ReasoningParser):
     text. This parser extracts all content before </think> as reasoning content.
     """
 
-    def __init__(self, tokenizer: PreTrainedTokenizerBase):
+    def __init__(self, tokenizer: PreTrainedTokenizerBase, *args, **kwargs):
-        super().__init__(tokenizer)
+        super().__init__(tokenizer, *args, **kwargs)
         self.think_end_token = "</think>"
 
         self.reasoning_regex = re.compile(rf"(.*?){self.think_end_token}",