Merge pull request #20593 from ryan-crabbe/perf/reuse-litellm-params

perf: reuse LiteLLM_Params

litellm/router.py

@@ -5562,7 +5562,7 @@ class Router:
                     return
                 else:
                     deployment_model_info = self.get_router_model_info(
-                        deployment=deployment_info.model_dump(),
+                        deployment=deployment_info,
                         received_model_name=model_group,
                     )
                     # get tpm/rpm from deployment info
@@ -6836,7 +6836,7 @@ class Router:
 
     @overload
     def get_router_model_info(
-        self, deployment: dict, received_model_name: str, id: None = None
+        self, deployment: Union[dict, "Deployment"], received_model_name: str, id: None = None
     ) -> ModelMapInfo:
         pass
 
@@ -6848,7 +6848,7 @@ class Router:
 
     def get_router_model_info(
         self,
-        deployment: Optional[dict],
+        deployment: Optional[Union[dict, "Deployment"]],
         received_model_name: str,
         id: Optional[str] = None,
     ) -> ModelMapInfo:
@@ -6868,7 +6868,7 @@ class Router:
         if id is not None:
             _deployment = self.get_deployment(model_id=id)
             if _deployment is not None:
-                deployment = _deployment.model_dump(exclude_none=True)
+                deployment = _deployment
 
         if deployment is None:
             raise ValueError("Deployment not found")
@@ -6880,10 +6880,22 @@ class Router:
 
         model = base_model
 
-        ## GET PROVIDER
+        ## GET PROVIDER - reuse LiteLLM_Params if already constructed
+        litellm_params_data = deployment.get("litellm_params")
+        litellm_params: LiteLLM_Params
+        if isinstance(litellm_params_data, LiteLLM_Params):
+            litellm_params = litellm_params_data
+        elif isinstance(litellm_params_data, dict) and "model" in litellm_params_data:
+            litellm_params = LiteLLM_Params(**litellm_params_data)
+        else:
+            raise ValueError(
+                f"Deployment missing valid litellm_params. "
+                f"Got: {type(litellm_params_data).__name__}, "
+                f"deployment_id: {deployment.get('model_info', {}).get('id', 'unknown')}"
+            )
         _model, custom_llm_provider, _, _ = litellm.get_llm_provider(
-            model=deployment.get("litellm_params", {}).get("model", ""),
-            litellm_params=LiteLLM_Params(**deployment.get("litellm_params", {})),
+            model=litellm_params.model,
+            litellm_params=litellm_params,
         )
 
         ## SET MODEL TO 'model=' - if base_model is None + not azure

tests/router_unit_tests/test_router_helper_utils.py

@@ -2198,3 +2198,33 @@ def test_get_valid_args():
     # Verify it contains keyword-only arguments too
     # These are common Router.__init__ parameters
     assert "assistants_config" in valid_args or "search_tools" in valid_args
+
+
+def test_get_router_model_info_with_deployment_object():
+    """Test get_router_model_info accepts Deployment object directly and reuses LiteLLM_Params"""
+    router = Router(
+        model_list=[
+            {
+                "model_name": "gpt-4",
+                "litellm_params": {"model": "gpt-4", "api_key": "test-key"},
+                "model_info": {"id": "test-id"},
+            }
+        ]
+    )
+
+    # Get the Deployment object (not dict)
+    deployment = router.get_deployment(model_id="test-id")
+    assert deployment is not None
+    assert isinstance(deployment, Deployment)
+    assert isinstance(deployment.litellm_params, LiteLLM_Params)
+
+    # Pass Deployment directly (not .model_dump()) - this exercises the isinstance check
+    # that reuses the existing LiteLLM_Params instead of reconstructing it
+    model_info = router.get_router_model_info(
+        deployment=deployment,
+        received_model_name="gpt-4",
+    )
+
+    # Verify we got valid model info back
+    assert model_info is not None
+    assert isinstance(model_info, dict)