feat: Add model version tracking with API endpoints and response metadata (#8795)

python/sglang/srt/entrypoints/openai/protocol.py

@@ -240,6 +240,7 @@ class CompletionResponse(BaseModel):
     model: str
     choices: List[CompletionResponseChoice]
     usage: UsageInfo
+    metadata: Optional[Dict[str, Any]] = None
 
 
 class CompletionResponseStreamChoice(BaseModel):
@@ -517,6 +518,7 @@ class ChatCompletionResponse(BaseModel):
     model: str
     choices: List[ChatCompletionResponseChoice]
     usage: UsageInfo
+    metadata: Optional[Dict[str, Any]] = None
 
 
 class DeltaMessage(BaseModel):

python/sglang/srt/entrypoints/openai/serving_chat.py

@@ -723,6 +723,7 @@ class OpenAIServingChat(OpenAIServingBase):
             model=request.model,
             choices=choices,
             usage=usage,
+            metadata={"weight_version": ret[0]["meta_info"]["weight_version"]},
         )
 
     def _process_logprobs_tokens(

python/sglang/srt/entrypoints/openai/serving_completions.py

@@ -373,6 +373,7 @@ class OpenAIServingCompletion(OpenAIServingBase):
             created=created,
             choices=choices,
             usage=usage,
+            metadata={"weight_version": ret[0]["meta_info"]["weight_version"]},
         )
 
     def _get_echo_text(self, request: CompletionRequest, index: int) -> str: