Fix tool_parser

Author: liuyanyi

vllm/entrypoints/openai/serving_chat.py

@@ -275,9 +275,6 @@ async def chat_completion_stream_generator(
 
         num_prompt_tokens = 0
 
-        tool_parser: Optional[ToolParser] = self.tool_parser(
-            tokenizer) if self.tool_parser else None
-
         if isinstance(request.tool_choice, ChatCompletionNamedToolChoiceParam):
             tool_choice_function_name = request.tool_choice.function.name
         else:
@@ -296,6 +293,19 @@ async def chat_completion_stream_generator(
         else:
             previous_texts, all_previous_token_ids = None, None
 
+        # Prepare the tool parser if it's needed
+        try:
+            if tool_choice_auto and self.tool_parser:
+                tool_parser: Optional[ToolParser] = self.tool_parser(tokenizer)
+            else:
+                tool_parser = None
+        except RuntimeError as e:
+            logger.error("Error in tool parser creation: %s", e)
+            data = self.create_streaming_error_response(str(e))
+            yield f"data: {data}\n\n"
+            yield "data: [DONE]\n\n"
+            return
+
         try:
             async for res in result_generator:
                 if res.prompt_token_ids is not None:
@@ -664,7 +674,12 @@ async def chat_completion_full_generator(
                     or request.tool_choice is None) and self.enable_auto_tools \
                     and self.tool_parser:
 
-                tool_parser = self.tool_parser(tokenizer)
+                try:
+                    tool_parser = self.tool_parser(tokenizer)
+                except RuntimeError as e:
+                    logger.error("Error in tool parser creation: %s", e)
+                    return self.create_error_response(str(e))
+
                 tool_call_info = tool_parser.extract_tool_calls(output.text)
                 tools_called = tool_call_info.tools_called
                 if tool_call_info.tools_called:

vllm/entrypoints/openai/tool_parsers/hermes_tool_parser.py

@@ -48,10 +48,10 @@ def __init__(self, tokenizer: AnyTokenizer):
             raise ValueError(
                 "The model tokenizer must be passed to the ToolParser "
                 "constructor during construction.")
-        self.tool_call_start_token_id: int = self.model_tokenizer.vocab[
-            self.tool_call_start_token]
-        self.tool_call_end_token_id: int = self.model_tokenizer.vocab[
-            self.tool_call_end_token]
+        self.tool_call_start_token_id: int = self.model_tokenizer.vocab.get(
+            self.tool_call_start_token, None)
+        self.tool_call_end_token_id: int = self.model_tokenizer.vocab.get(
+            self.tool_call_end_token, None)
         if not self.tool_call_start_token_id or not self.tool_call_end_token_id:
             raise RuntimeError(
                 "Hermes 2 Pro Tool parser could not locate tool call start/end "

vllm/entrypoints/openai/tool_parsers/mistral_tool_parser.py

@@ -45,8 +45,13 @@ def __init__(self, tokenizer: AnyTokenizer):
         self.streamed_args_for_tool: List[str] = [
         ]  # map what has been streamed for each tool so far to a list
         self.bot_token = "[TOOL_CALLS]"
-        self.bot_token_id = self.model_tokenizer.vocab[self.bot_token]
+        self.bot_token_id = self.model_tokenizer.vocab.get(
+            self.bot_token, None)
         self.tool_call_regex = re.compile(r"\[{.*?}\]", re.DOTALL)
+        if not self.bot_token_id:
+            raise RuntimeError(
+                "Mistral Tool Parser could not locate the tool call token in "
+                "the tokenizer!")
 
     def extract_tool_calls(self,
                            model_output: str) -> ExtractedToolCallInformation: