fix(proxy): make async_pre_call_hook str returns reach the caller as completion content

process_pre_call_hook_response only routed "completion"/"text_completion" onto the RejectedRequestError path, but /v1/chat/completions and /v1/completions run with call_type "acompletion"/"atext_completion", so a documented str return 400'd instead.

RejectedRequestError now keeps the unprefixed text in raw_message and the proxy uses it for completion content, so callers no longer see "litellm.BadRequestError: litellm.RejectedRequestError: ..." in the message body.

The streaming branch also crashed with AttributeError because post_call_failure_hook pops litellm_logging_obj before CustomStreamWrapper is built; capture it first.

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

litellm/exceptions.py

@@ -553,6 +553,7 @@ class RejectedRequestError(BadRequestError):  # type: ignore
         litellm_debug_info: Optional[str] = None,
     ):
         self.status_code = 400
+        self.raw_message = message
         self.message = "litellm.RejectedRequestError: {}".format(message)
         self.model = model
         self.llm_provider = llm_provider

litellm/proxy/proxy_server.py

@@ -9172,13 +9172,14 @@ async def chat_completion(
         return _chat_response
     except RejectedRequestError as e:
         _data = e.request_data
+        _logging_obj = _data.get("litellm_logging_obj")
         await proxy_logging_obj.post_call_failure_hook(
             user_api_key_dict=user_api_key_dict,
             original_exception=e,
             request_data=_data,
         )
         _chat_response = litellm.ModelResponse()
-        _chat_response.choices[0].message.content = e.message  # type: ignore
+        _chat_response.choices[0].message.content = e.raw_message  # type: ignore
 
         if data.get("stream", None) is not None and data["stream"] is True:
             _iterator = litellm.utils.ModelResponseIterator(model_response=_chat_response, convert_to_delta=True)
@@ -9186,7 +9187,7 @@ async def chat_completion(
                 completion_stream=_iterator,
                 model=data.get("model", ""),
                 custom_llm_provider="cached_response",
-                logging_obj=_data.get("litellm_logging_obj", None),
+                logging_obj=_logging_obj,
             )
             selected_data_generator = select_data_generator(
                 response=_streaming_response,
@@ -9340,7 +9341,7 @@ async def completion(
                 total_tokens=0,
             )
             _chat_response.usage = _usage  # type: ignore
-            _chat_response.choices[0].message.content = e.message  # type: ignore
+            _chat_response.choices[0].message.content = e.raw_message  # type: ignore
             _iterator = litellm.utils.ModelResponseIterator(model_response=_chat_response, convert_to_delta=True)
             _streaming_response = litellm.TextCompletionStreamWrapper(
                 completion_stream=_iterator,
@@ -9362,7 +9363,7 @@ async def completion(
             )
         else:
             _response = litellm.TextCompletionResponse()
-            _response.choices[0].text = e.message
+            _response.choices[0].text = e.raw_message
             return _response
     except Exception as e:
         await proxy_logging_obj.post_call_failure_hook(

litellm/proxy/utils.py

@@ -189,6 +189,16 @@ unified_guardrail = UnifiedLLMGuardrails()
 
 NON_OPENAI_STREAM_GUARDRAIL_TRANSLATION_CALL_TYPES: "frozenset[CallTypes]" = frozenset({CallTypes.anthropic_messages})
 
+REJECTED_REQUEST_SUPPORTED_CALL_TYPES: "frozenset[str]" = frozenset(
+    call_type.value
+    for call_type in (
+        CallTypes.completion,
+        CallTypes.acompletion,
+        CallTypes.text_completion,
+        CallTypes.atext_completion,
+    )
+)
+
 
 def print_verbose(print_statement):
     """
@@ -920,7 +930,7 @@ class ProxyLogging:
         if isinstance(response, dict):
             return response
         if isinstance(response, str):
-            if call_type in ["completion", "text_completion"]:
+            if call_type in REJECTED_REQUEST_SUPPORTED_CALL_TYPES:
                 raise RejectedRequestError(
                     message=response,
                     model=data.get("model", ""),

tests/test_litellm/proxy/utils/proxy_logging/test_pre_call_hook.py

@@ -46,6 +46,18 @@ async def test_process_pre_call_hook_response_string_completion_raises_rejected(
         )
 
 
+@pytest.mark.parametrize("call_type", ["completion", "acompletion", "text_completion", "atext_completion"])
+@pytest.mark.asyncio
+async def test_process_pre_call_hook_response_string_completion_call_types_raise_rejected(proxy_logging, call_type):
+    with pytest.raises(RejectedRequestError) as info:
+        await proxy_logging.process_pre_call_hook_response(
+            response="rejected",
+            data={"model": "m"},
+            call_type=call_type,
+        )
+    assert info.value.raw_message == "rejected"
+
+
 @pytest.mark.asyncio
 async def test_process_pre_call_hook_response_string_other_call_type_raises_http(proxy_logging):
     with pytest.raises(HTTPException) as info:

tests/test_litellm/test_rejected_request_error_message.py

@@ -0,0 +1,11 @@
+"""Regression: ``RejectedRequestError`` keeps the unprefixed text for completion content."""
+
+from __future__ import annotations
+
+from litellm.exceptions import RejectedRequestError
+
+
+def test_rejected_request_error_keeps_raw_message():
+    err = RejectedRequestError(message="366", model="gpt-4o", llm_provider="", request_data={"model": "gpt-4o"})
+    assert err.raw_message == "366"
+    assert "RejectedRequestError" in err.message