[Frontend][Core] Move guided decoding params into sampling params (#8252)

Signed-off-by: Joe Runde <Joseph.Runde@ibm.com>
Co-authored-by: Nick Hill <nickhill@us.ibm.com>

vllm/entrypoints/openai/serving_chat.py

@@ -187,9 +187,6 @@ class OpenAIServingChat(OpenAIServing):
             raw_request.state.request_metadata = request_metadata
 
         try:
-            guided_decode_logits_processor = (
-                await self._guided_decode_logits_processor(request, tokenizer))
-
             if isinstance(prompt, str):
                 prompt_inputs = self._tokenize_prompt_input(
                     request,
@@ -208,8 +205,6 @@ class OpenAIServingChat(OpenAIServing):
             assert prompt_inputs is not None
 
             sampling_params = request.to_sampling_params(
-                tokenizer,
-                guided_decode_logits_processor,
                 default_max_tokens=self.max_model_len -
                 len(prompt_inputs["prompt_token_ids"]))