diff --git a/tests/entrypoints/openai/parser/test_harmony_utils.py b/tests/entrypoints/openai/parser/test_harmony_utils.py index 0027c2763f..1b19714ac2 100644 --- a/tests/entrypoints/openai/parser/test_harmony_utils.py +++ b/tests/entrypoints/openai/parser/test_harmony_utils.py @@ -20,6 +20,7 @@ from vllm.entrypoints.openai.responses.harmony import ( response_input_to_harmony, response_previous_input_to_harmony, ) +from vllm.exceptions import VLLMValidationError _TOOL_PARAMETERS = { "type": "object", @@ -975,10 +976,11 @@ class TestGetSystemMessage: def test_unsupported_reasoning_effort_raises_clear_error(self) -> None: with pytest.raises( - ValueError, + VLLMValidationError, match="reasoning_effort='max' is not supported by Harmony", - ): + ) as exc_info: get_system_message(reasoning_effort="max") + assert exc_info.value.parameter == "reasoning_effort" class TestResponseInputToHarmonyReasoningItem: diff --git a/tests/entrypoints/openai/responses/test_responses_utils.py b/tests/entrypoints/openai/responses/test_responses_utils.py index 2cda73d98a..967ddacda6 100644 --- a/tests/entrypoints/openai/responses/test_responses_utils.py +++ b/tests/entrypoints/openai/responses/test_responses_utils.py @@ -22,6 +22,7 @@ from vllm.entrypoints.openai.responses.utils import ( construct_input_messages, should_continue_final_message, ) +from vllm.exceptions import VLLMValidationError def _single_chat_message(item): @@ -223,8 +224,9 @@ class TestResponsesUtils: encrypted_content="TOP_SECRET_MESSAGE", status=None, ) - with pytest.raises(ValueError): + with pytest.raises(VLLMValidationError) as exc_info: construct_chat_messages_with_tool_call([item]) + assert exc_info.value.parameter == "input" output_item = ResponseOutputMessage( id="msg_bf585bbbe3d500e0", @@ -350,7 +352,7 @@ class TestReasoningItemContentPriority: assert formatted["reasoning"] == "" def test_encrypted_content_raises(self): - """Encrypted content should still raise ValueError.""" + """Encrypted content should raise VLLMValidationError.""" item = ResponseReasoningItem( id="reasoning_6", summary=[ @@ -369,8 +371,9 @@ class TestReasoningItemContentPriority: encrypted_content="ENCRYPTED", status=None, ) - with pytest.raises(ValueError): + with pytest.raises(VLLMValidationError) as exc_info: construct_chat_messages_with_tool_call([item]) + assert exc_info.value.parameter == "input" @patch("vllm.entrypoints.openai.responses.utils.logger") def test_summary_with_multiple_entries_uses_first(self, mock_logger): diff --git a/vllm/entrypoints/openai/parser/harmony_utils.py b/vllm/entrypoints/openai/parser/harmony_utils.py index a18a93704c..fca02a91cc 100644 --- a/vllm/entrypoints/openai/parser/harmony_utils.py +++ b/vllm/entrypoints/openai/parser/harmony_utils.py @@ -24,6 +24,7 @@ from openai_harmony import ( from vllm import envs from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionToolsParam +from vllm.exceptions import VLLMValidationError from vllm.logger import init_logger logger = init_logger(__name__) @@ -123,9 +124,10 @@ def get_system_message( if reasoning_effort is not None: if reasoning_effort not in REASONING_EFFORT: supported_values = ", ".join(REASONING_EFFORT) - raise ValueError( + raise VLLMValidationError( f"reasoning_effort={reasoning_effort!r} is not supported by " - f"Harmony. Supported values are: {supported_values}." + f"Harmony. Supported values are: {supported_values}.", + parameter="reasoning_effort", ) sys_msg_content = sys_msg_content.with_reasoning_effort( REASONING_EFFORT[reasoning_effort] @@ -181,7 +183,10 @@ def get_developer_message( elif tool.type == "function": function_tools.append(tool) else: - raise ValueError(f"tool type {tool.type} not supported") + raise VLLMValidationError( + f"tool type {tool.type!r} is not supported.", + parameter="tools", + ) if function_tools: function_tool_descriptions = [ create_tool_definition(tool) for tool in function_tools @@ -297,7 +302,10 @@ def extract_instructions_from_messages( elif hasattr(first_message, "model_dump"): first_message = first_message.model_dump(exclude_none=True) else: - raise ValueError(f"Unknown message type: {type(first_message)}") + raise VLLMValidationError( + f"Unknown message type: {type(first_message)}", + parameter="input", + ) if first_message.get("role") not in ( "system", diff --git a/vllm/entrypoints/openai/responses/utils.py b/vllm/entrypoints/openai/responses/utils.py index 686c1fa292..173eb37906 100644 --- a/vllm/entrypoints/openai/responses/utils.py +++ b/vllm/entrypoints/openai/responses/utils.py @@ -37,6 +37,7 @@ from vllm.entrypoints.openai.chat_completion.protocol import ( ChatCompletionToolsParam, ) from vllm.entrypoints.openai.responses.protocol import ResponseInputOutputItem +from vllm.exceptions import VLLMValidationError from vllm.logger import init_logger from vllm.tool_parsers.utils import ( build_responses_tool_call_name_map, @@ -273,7 +274,10 @@ def _construct_message_from_response_item( elif isinstance(item, ResponseReasoningItem): reasoning = "" if item.encrypted_content: - raise ValueError("Encrypted content is not supported.") + raise VLLMValidationError( + "Encrypted content is not supported.", + parameter="input", + ) elif item.content and len(item.content) >= 1: reasoning = item.content[0].text elif len(item.summary) >= 1: