support vllm==0.4.2

Author: FreeRotate

api/utils/protocol.py

@@ -236,6 +236,8 @@ class ChatCompletionCreateParams(BaseModel):
 
     guided_grammar: Optional[str] = None
 
+    guided_decoding_backend: Optional[str] = "lm-format-enforcer"
+
 
 class CompletionCreateParams(BaseModel):
     model: str

api/vllm_routes/chat.py

@@ -106,6 +106,7 @@ async def create_chat_completion(
             from vllm.model_executor.guided_decoding import get_guided_decoding_logits_processor
             guided_decode_logits_processor = (
                 await get_guided_decoding_logits_processor(
+                    request.guided_decoding_backend,
                     request,
                     engine.tokenizer,
                 )

api/vllm_routes/completion.py

@@ -145,6 +145,7 @@ async def create_completion(
             from vllm.model_executor.guided_decoding import get_guided_decoding_logits_processor
             guided_decode_logits_processor = (
                 await get_guided_decoding_logits_processor(
+                    request.guided_decoding_backend,
                     request,
                     engine.tokenizer,
                 )