[Fix] Fix Structured output Performance issues (#69)

Fix Structured output Performance issues

Author: psriramsnc

sygra/core/models/custom_models.py

@@ -221,30 +221,27 @@ async def _handle_structured_output(
         self, input: ChatPromptValue, model_params: ModelParams, **kwargs: Any
     ) -> Optional[ModelResponse]:
         """Handle structured output generation"""
-        lock = await self._get_lock()
-        # Re-entry prevention using asyncio locking
-        async with lock:
-            pydantic_model = self.structured_output.get_pydantic_model()
-            if not pydantic_model:
-                logger.warning(
-                    "Structured output enabled but no valid schema found, falling back to regular generation"
-                )
-                # Return a flag to signal that regular generation should be used
-                return None
+        pydantic_model = self.structured_output.get_pydantic_model()
+        if not pydantic_model:
+            logger.warning(
+                "Structured output enabled but no valid schema found, falling back to regular generation"
+            )
+            # Return a flag to signal that regular generation should be used
+            return None
 
-            # Check if model supports native structured output
-            if self._supports_native_structured_output():
-                logger.info(f"Using native structured output for {self.name()}")
-                return await self._generate_native_structured_output(
-                    input, model_params, pydantic_model, **kwargs
-                )
-            else:
-                logger.info(f"Using fallback structured output for {self.name()}")
-                # Get response from fallback method
-                model_response: ModelResponse = await self._generate_fallback_structured_output(
-                    input, model_params, pydantic_model, **kwargs
-                )
-                return model_response
+        # Check if model supports native structured output
+        if self._supports_native_structured_output():
+            logger.info(f"Using native structured output for {self.name()}")
+            return await self._generate_native_structured_output(
+                input, model_params, pydantic_model, **kwargs
+            )
+        else:
+            logger.info(f"Using fallback structured output for {self.name()}")
+            # Get response from fallback method
+            model_response: ModelResponse = await self._generate_fallback_structured_output(
+                input, model_params, pydantic_model, **kwargs
+            )
+            return model_response
 
     def _supports_native_structured_output(self) -> bool:
         """Check if the model supports native structured output"""

sygra/core/models/lite_llm/vllm_model.py

@@ -22,7 +22,7 @@ def __init__(self, model_config: dict[str, Any]) -> None:
         utils.validate_required_keys(["url", "auth_token"], model_config, "model")
         self.model_config = model_config
         self.auth_token = str(model_config.get("auth_token")).replace("Bearer ", "")
-        self.model_name = model_config.get("model_serving_name", self.name())
+        self.model_name = model_config.get("model_serving_name", self.model_name)
 
     def _validate_completions_api_model_support(self) -> None:
         logger.info(f"Model {self.name()} supports completion API.")