refactor: type fresh proxy cache and routing hooks, drop private-import suppression

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/litellm_core_utils/prompt_templates/common_utils.py

@@ -179,15 +179,15 @@ from litellm.litellm_core_utils.prompt_templates.tool_call_json_repair import (
 from litellm.litellm_core_utils.prompt_templates.tool_call_json_repair import (
     split_concatenated_json_objects as split_concatenated_json_objects,  # noqa: PLC0414  # public compatibility re-export
 )
-from litellm.litellm_core_utils.prompt_templates.tpm_media_utils import (
-    _image_part_url,  # pyright: ignore[reportPrivateUsage]  # Shared implementation for as_openai_image_part
-)
 from litellm.litellm_core_utils.prompt_templates.tpm_media_utils import (
     anthropic_image_source_to_openai_url as anthropic_image_source_to_openai_url,  # noqa: PLC0414  # public compatibility re-export
 )
 from litellm.litellm_core_utils.prompt_templates.tpm_media_utils import (
     embedding_input_is_all_media as embedding_input_is_all_media,  # noqa: PLC0414  # public compatibility re-export
 )
+from litellm.litellm_core_utils.prompt_templates.tpm_media_utils import (
+    image_part_url,
+)
 from litellm.litellm_core_utils.prompt_templates.tpm_media_utils import (
     split_embedding_media_for_tpm as split_embedding_media_for_tpm,  # noqa: PLC0414  # public compatibility re-export
 )
@@ -537,7 +537,7 @@ def as_openai_image_part(part: Mapping[str, object]) -> ChatCompletionImageObjec
 
     Rebuilt rather than forwarded so no caller-controlled key beyond the URL rides along.
     """
-    url: Final = _image_part_url(part)
+    url: Final = image_part_url(part)
     return {"type": "image_url", "image_url": {"url": url}} if url else None
 
 

litellm/litellm_core_utils/prompt_templates/tpm_media_utils.py

@@ -83,7 +83,7 @@ def _tpm_media_leaf(value: object) -> _TPMMediaProjection:
         image: Final = as_str_mapping(part.get("image_url"))
         source: Final = as_str_mapping(part.get("source"))
         canonical: Final = _tpm_image_part(
-            _image_part_url({**part, "type": "input_image"} if kind == "computer_screenshot" else part) or "",
+            image_part_url({**part, "type": "input_image"} if kind == "computer_screenshot" else part) or "",
             (image or part).get("detail", part.get("detail", "auto")),
         )
         projected: Final = (
@@ -243,7 +243,7 @@ def anthropic_image_source_to_openai_url(image_source: Mapping[str, object]) ->
     return None
 
 
-def _image_part_url(part: Mapping[str, object]) -> str | None:
+def image_part_url(part: Mapping[str, object]) -> str | None:
     """The image URL carried by one content part, whichever of the three dialects wrote it."""
     part_type: Final = part.get("type")
     if part_type in ("image_url", "input_image"):

litellm/proxy/common_utils/user_api_key_cache.py

@@ -289,7 +289,7 @@ class UserApiKeyCache(DualCache):
             verbose_proxy_logger.error("UserApiKeyCache.async_get_cache failed for key=%r: %s", key, e)
             return None
 
-    def set_cache(self, key: str | None, value: CacheValue, local_only: bool = False, **kwargs: object):
+    def set_cache(self, key: str | None, value: CacheValue, local_only: bool = False, **kwargs: object) -> None:
         model_type: Final = cast(type[BaseModel] | None, kwargs.pop("model_type", None))
         payload: Final[object] = CacheCodec.serialize(value, model_type=model_type)
         shared: Final = self._shared
@@ -308,7 +308,9 @@ class UserApiKeyCache(DualCache):
             return self.key_object_cache.set_cache(key=key, value=payload, local_only=local_only, **kwargs)
         return super().set_cache(key=key, value=payload, local_only=local_only, **kwargs)
 
-    async def async_set_cache(self, key: str | None, value: CacheValue, local_only: bool = False, **kwargs: object):
+    async def async_set_cache(
+        self, key: str | None, value: CacheValue, local_only: bool = False, **kwargs: object
+    ) -> None:
         """Inside a request the Redis SET rides the request's pipeline (memory is written at once); anywhere
         else, or with options the pipeline does not carry, it goes to Redis directly as before."""
         model_type: Final = cast(type[BaseModel] | None, kwargs.pop("model_type", None))

litellm/proxy/hooks/sensitive_data_routing.py

@@ -41,7 +41,9 @@ class _PROXY_SensitiveDataRoutingHandler(CustomLogger):
     model field if a routing override exists for the session.
     """
 
-    def __init__(self, internal_usage_cache: InternalUsageCache, route_pins: SensitiveRoutePinStore | None = None):
+    def __init__(
+        self, internal_usage_cache: InternalUsageCache, route_pins: SensitiveRoutePinStore | None = None
+    ) -> None:
         self.route_pins: Final = route_pins or SensitiveRoutePinStore(
             connect(internal_usage_cache.dual_cache.redis_cache).sensitive_route_pins()
         )