fix(typing): keep router fence visibility and py310-safe Self import

Bind the Protocol-narrowed routers in pass_through_endpoints and rag
ingestion to router-named locals so router_surface_check still sees the
aadapter_completion and aocr uses, and import Self from typing_extensions
in prisma_client for check_py310_typing_imports.

litellm/proxy/db/prisma_client.py

@@ -37,11 +37,11 @@ from litellm.secret_managers.main import str_to_bool
 if TYPE_CHECKING:
     from collections.abc import Sequence
     from types import TracebackType
-    from typing import Self
 
     from prisma import Prisma
     from prisma import models as prisma_models
     from prisma.types import HttpConfig
+    from typing_extensions import Self
 
     from litellm.repositories.prisma_protocols import TableActions
 

litellm/proxy/pass_through_endpoints/pass_through_endpoints.py

@@ -374,11 +374,8 @@ async def chat_completion_pass_through_endpoint(
             and data["model"] is not None
             and llm_router.registry().resolve(data["model"], kind="recognized")
         ):
-            llm_response = asyncio.create_task(
-                require_attributes(llm_router, _AdapterCompletionRouter, "aadapter_completion").aadapter_completion(
-                    **data
-                )
-            )
+            adapter_router = require_attributes(llm_router, _AdapterCompletionRouter, "aadapter_completion")
+            llm_response = asyncio.create_task(adapter_router.aadapter_completion(**data))
         elif (
             llm_router is not None
             and data["model"] not in router_model_names
@@ -387,19 +384,13 @@ async def chat_completion_pass_through_endpoint(
                 or len(llm_router.registry().patterns().patterns) > 0
             )
         ):  # check for wildcard routes or default deployment before checking deployment_names
-            llm_response = asyncio.create_task(
-                require_attributes(llm_router, _AdapterCompletionRouter, "aadapter_completion").aadapter_completion(
-                    **data
-                )
-            )
+            adapter_router = require_attributes(llm_router, _AdapterCompletionRouter, "aadapter_completion")
+            llm_response = asyncio.create_task(adapter_router.aadapter_completion(**data))
         elif llm_router is not None and data["model"] in llm_router.registry().names(
             kind="deployment_names"
         ):  # model in router deployments, calling a specific deployment on the router (lowest priority)
-            llm_response = asyncio.create_task(
-                require_attributes(llm_router, _AdapterCompletionRouter, "aadapter_completion").aadapter_completion(
-                    **data, specific_deployment=True
-                )
-            )
+            adapter_router = require_attributes(llm_router, _AdapterCompletionRouter, "aadapter_completion")
+            llm_response = asyncio.create_task(adapter_router.aadapter_completion(**data, specific_deployment=True))
         elif user_model is not None:  # `litellm --model <your-model-name>`
             llm_response = asyncio.create_task(litellm.aadapter_completion(**data))
         else:

litellm/rag/ingestion/base_ingestion.py

@@ -186,7 +186,8 @@ class BaseRAGIngestion(ABC):
 
         # Use router if available
         if self.router is not None:
-            ocr_response = await require_attributes(self.router, _OcrRouter, "aocr").aocr(
+            ocr_router: Final = require_attributes(self.router, _OcrRouter, "aocr")
+            ocr_response = await ocr_router.aocr(
                 model=ocr_model,
                 document={"type": doc_type, url_key: data_url},
             )