fix(proxy): type RejectedRequestError.model and read it directly

The type check gate counted four new reportUnknownMemberType errors from
reading the untyped request_data dict and model attribute. Annotate model
as str, like ModifyResponseException, and use it directly; the
request-data fallback only covered callers passing model=None, which
the annotation now rules out.

litellm/exceptions.py

@@ -637,7 +637,7 @@ class RejectedRequestError(BadRequestError):
     def __init__(
         self,
         message,
-        model,
+        model: str,
         llm_provider,
         request_data: dict,
         litellm_debug_info: str | None = None,

litellm/proxy/proxy_server.py

@@ -12824,7 +12824,7 @@ async def chat_completion(
             request_data=_data,
         )
         _chat_response = litellm.ModelResponse()
-        _chat_response.model = _data.get("model") or e.model or ""
+        _chat_response.model = e.model
         _chat_response.choices[0].message.content = e.message
 
         if data.get("stream", None) is not None and data["stream"] is True:
@@ -13010,7 +13010,7 @@ async def completion(
         else:
             _response = litellm.TextCompletionResponse()
             _response.choices[0].text = e.message
-            _response.model = _data.get("model") or e.model or ""
+            _response.model = e.model
             return _response
     except Exception as e:
         await proxy_logging_obj.post_call_failure_hook(

tests/unit/proxy/test_rejected_request_response_model.py

@@ -17,10 +17,10 @@ from litellm.proxy._types import UserAPIKeyAuth
 from litellm.proxy.proxy_server import chat_completion, completion
 
 
-async def _run_rejected_request(route, body, exception_model="my-model"):
+async def _run_rejected_request(route, body):
     exception = RejectedRequestError(
         message="cached answer",
-        model=exception_model,
+        model=body["model"],
         llm_provider="",
         request_data=dict(body),
     )
@@ -64,18 +64,3 @@ async def test_rejected_text_completion_response_has_model():
     # finish_reason is still None here, so this is not validated against
     # openai.types.Completion the way the chat response is.
     assert response.model == "my-model"
-
-
-@pytest.mark.asyncio
-@pytest.mark.parametrize(
-    "route, body",
-    [
-        (chat_completion, {"model": "my-model", "messages": [{"role": "user", "content": "hi"}]}),
-        (completion, {"model": "my-model", "prompt": "hi"}),
-    ],
-)
-async def test_rejected_response_falls_back_to_request_model(route, body):
-    # RejectedRequestError is public, so a custom hook can raise it without a model.
-    response = await _run_rejected_request(route, body, exception_model=None)
-
-    assert response.model == "my-model"