fix(llm/clients): do not append 0.21 hint when stream_options guidance fires

Author: Pouyanpi

nemoguardrails/llm/clients/_errors.py

@@ -173,7 +173,8 @@ def _classify_bad_request(status_code: int, error_message: str, kwargs: Dict[str
                 f"{error_message} (set include_usage_in_stream=False on the model "
                 "or in config.yml parameters to remove this field from streaming requests)"
             )
-        error_message = _maybe_append_migration_hint(error_message)
+        else:
+            error_message = _maybe_append_migration_hint(error_message)
         return LLMUnsupportedParamsError(status_code, error_message, **kwargs)
     return LLMBadRequestError(status_code, error_message, **kwargs)
 

tests/llm/clients/test_openai_compatible_400_enrichment.py

@@ -165,6 +165,20 @@ async def test_unrecognized_auth_scheme_no_hint(self):
         assert _HINT_FRAGMENT not in exc_info.value.error_message
 
 
+class TestStreamOptionsRejection:
+    @pytest.mark.asyncio
+    async def test_stream_options_rejection_uses_specific_guidance_only(self):
+        client = make_client()
+        mock_httpx_post(
+            client,
+            [(400, {"error": {"message": "Unrecognized request argument supplied: stream_options"}}, {})],
+        )
+        with pytest.raises(LLMUnsupportedParamsError) as exc_info:
+            await client.chat_completion("gpt-4o", [])
+        assert "include_usage_in_stream=False" in exc_info.value.error_message
+        assert _HINT_FRAGMENT not in exc_info.value.error_message
+
+
 class TestMigrationHintNotAppendedOnUnrelated400:
     @pytest.mark.asyncio
     async def test_generic_400_no_hint(self):