From 18101345a015540c3b5643f2e0c4206c54700f72 Mon Sep 17 00:00:00 2001 From: Elliott de Launay Date: Tue, 7 Jul 2026 17:30:19 -0400 Subject: [PATCH] fix(responses): propagate message cache_control safely through objects and models --- .../context_caching/transformation.py | 22 +- .../transformation.py | 65 +- litellm/utils.py | 53 +- .../test_litellm_completion_responses.py | 559 ++++++++---------- 4 files changed, 348 insertions(+), 351 deletions(-) diff --git a/litellm/llms/vertex_ai/context_caching/transformation.py b/litellm/llms/vertex_ai/context_caching/transformation.py index f0ce3323ef6..183d9743f13 100644 --- a/litellm/llms/vertex_ai/context_caching/transformation.py +++ b/litellm/llms/vertex_ai/context_caching/transformation.py @@ -62,23 +62,27 @@ def extract_ttl_from_cached_messages(messages: List[AllMessageValues]) -> Option if not is_cached_message(message): continue - content = message.get("content") + content = message.get("content") if isinstance(message, dict) else getattr(message, "content", None) if not content or isinstance(content, str): continue for content_item in content: - # Type check to ensure content_item is a dictionary before calling .get() - if not isinstance(content_item, dict): + # Check if content_item is dict or object model + if isinstance(content_item, dict): + cache_control = content_item.get("cache_control") + else: + cache_control = getattr(content_item, "cache_control", None) + + if not cache_control: continue - cache_control = content_item.get("cache_control") - if not cache_control or not isinstance(cache_control, dict): + cc_type = ( + cache_control.get("type") if isinstance(cache_control, dict) else getattr(cache_control, "type", None) + ) + if cc_type != "ephemeral": continue - if cache_control.get("type") != "ephemeral": - continue - - ttl = cache_control.get("ttl") + ttl = cache_control.get("ttl") if isinstance(cache_control, dict) else getattr(cache_control, "ttl", None) if ttl and _is_valid_ttl_format(ttl): return str(ttl) diff --git a/litellm/responses/litellm_completion_transformation/transformation.py b/litellm/responses/litellm_completion_transformation/transformation.py index 6b1ca3564e3..8052f69f22b 100644 --- a/litellm/responses/litellm_completion_transformation/transformation.py +++ b/litellm/responses/litellm_completion_transformation/transformation.py @@ -892,26 +892,38 @@ class LiteLLMCompletionResponsesConfig: function_call=input_item ) else: - content = input_item.get("content") + content = ( + input_item.get("content") if isinstance(input_item, dict) else getattr(input_item, "content", None) + ) # Handle None content: Responses API allows None content, but GenericChatCompletionMessage requires content # Since guardrails skip None content anyway, we return empty list to exclude it from structured messages if content is None: return [] - return [ - GenericChatCompletionMessage( - role=input_item.get("role") or "user", - content=LiteLLMCompletionResponsesConfig._transform_responses_api_content_to_chat_completion_content( - content - ), - ) - ] + + role = input_item.get("role") if isinstance(input_item, dict) else getattr(input_item, "role", None) + cache_control = ( + input_item.get("cache_control") + if isinstance(input_item, dict) + else getattr(input_item, "cache_control", None) + ) + + msg = GenericChatCompletionMessage( + role=role or "user", + content=LiteLLMCompletionResponsesConfig._transform_responses_api_content_to_chat_completion_content( + content + ), + ) + if cache_control is not None: + msg["cache_control"] = cache_control + return [msg] @staticmethod def _is_input_item_tool_call_output(input_item: Any) -> bool: """ Check if the input item is a tool call output """ - return input_item.get("type") in [ + val = input_item.get("type") if isinstance(input_item, dict) else getattr(input_item, "type", None) + return val in [ "function_call_output", "custom_tool_call_output", "web_search_call", @@ -926,7 +938,8 @@ class LiteLLMCompletionResponsesConfig: Both need to be reconstructed as assistant tool_calls for Chat Completions providers. """ - return input_item.get("type") in ("function_call", "custom_tool_call") + val = input_item.get("type") if isinstance(input_item, dict) else getattr(input_item, "type", None) + return val in ("function_call", "custom_tool_call") @staticmethod def _transform_responses_api_tool_call_output_to_chat_completion_message( @@ -1155,6 +1168,31 @@ class LiteLLMCompletionResponsesConfig: return ChatCompletionImageObject(type="image_url", image_url=image_url_obj) + @staticmethod + def _normalize_responses_api_object_to_dict(item: Any) -> dict[str, Any]: + """ + Normalize a Responses API object (Pydantic model or custom class) to a dictionary + """ + if hasattr(item, "model_dump"): + return item.model_dump() + elif hasattr(item, "dict"): + return item.dict() + + item_dict = {} + for attr in [ + "type", + "text", + "cache_control", + "file_id", + "file_data", + "file_url", + "image_url", + "detail", + ]: + if hasattr(item, attr): + item_dict[attr] = getattr(item, attr) + return item_dict + @staticmethod def _transform_responses_api_content_to_chat_completion_content( content: Any, @@ -1176,7 +1214,10 @@ class LiteLLMCompletionResponsesConfig: for item in content: if isinstance(item, str): content_list.append(item) - elif isinstance(item, dict): + elif isinstance(item, dict) or (item is not None and not isinstance(item, (str, int, float, bool))): + if not isinstance(item, dict): + item = LiteLLMCompletionResponsesConfig._normalize_responses_api_object_to_dict(item) + if item.get("type") == "input_file": content_list.append( LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item(item) diff --git a/litellm/utils.py b/litellm/utils.py index a11c5500503..3fdf0a5eee6 100644 --- a/litellm/utils.py +++ b/litellm/utils.py @@ -7209,36 +7209,51 @@ def is_cached_message(message: AllMessageValues) -> bool: return False # Check message-level cache_control (set by cache_control_injection_points hook for string content) - message_level_cache_control = message.get("cache_control") - if ( - message_level_cache_control is not None - and isinstance(message_level_cache_control, dict) - and message_level_cache_control.get("type") == "ephemeral" - ): - return True + message_level_cache_control = ( + message.get("cache_control") + if isinstance(message, dict) + else getattr(message, "cache_control", None) + ) + if message_level_cache_control is not None: + cc_type = ( + message_level_cache_control.get("type") + if isinstance(message_level_cache_control, dict) + else getattr(message_level_cache_control, "type", None) + ) + if cc_type == "ephemeral": + return True - if "content" not in message: - return False - - content = message["content"] + if isinstance(message, dict): + if "content" not in message: + return False + content = message["content"] + else: + content = getattr(message, "content", None) # Handle non-list content types (None, str, etc.) if not isinstance(content, list): return False for content_item in content: - # Ensure content_item is a dictionary before accessing keys - if not isinstance(content_item, dict): - continue + # Check if content_item is dict or object model + if isinstance(content_item, dict): + cache_control = content_item.get("cache_control") + item_type = content_item.get("type") + else: + cache_control = getattr(content_item, "cache_control", None) + item_type = getattr(content_item, "type", None) - cache_control = content_item.get("cache_control") if ( - content_item.get("type") == "text" + item_type == "text" and cache_control is not None - and isinstance(cache_control, dict) - and cache_control.get("type") == "ephemeral" ): - return True + cc_type = ( + cache_control.get("type") + if isinstance(cache_control, dict) + else getattr(cache_control, "type", None) + ) + if cc_type == "ephemeral": + return True return False diff --git a/tests/test_litellm/responses/litellm_completion_transformation/test_litellm_completion_responses.py b/tests/test_litellm/responses/litellm_completion_transformation/test_litellm_completion_responses.py index d8e3f495ced..4b07d638261 100644 --- a/tests/test_litellm/responses/litellm_completion_transformation/test_litellm_completion_responses.py +++ b/tests/test_litellm/responses/litellm_completion_transformation/test_litellm_completion_responses.py @@ -3,9 +3,7 @@ import sys import pytest -sys.path.insert( - 0, os.path.abspath("../../..") -) # Adds the parent directory to the system path +sys.path.insert(0, os.path.abspath("../../..")) # Adds the parent directory to the system path from litellm.responses.litellm_completion_transformation.transformation import ( TOOL_CALLS_CACHE, @@ -34,11 +32,7 @@ class TestLiteLLMCompletionResponsesConfig: input_item = {"type": "input_file", "file_id": "file-abc123xyz"} # Execute - result = ( - LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item( - input_item - ) - ) + result = LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item(input_item) # Assert expected = {"type": "file", "file": {"file_id": "file-abc123xyz"}} @@ -53,11 +47,7 @@ class TestLiteLLMCompletionResponsesConfig: input_item = {"type": "input_file", "file_data": file_data} # Execute - result = ( - LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item( - input_item - ) - ) + result = LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item(input_item) # Assert expected = {"type": "file", "file": {"file_data": file_data}} @@ -75,11 +65,7 @@ class TestLiteLLMCompletionResponsesConfig: } # Execute - result = ( - LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item( - input_item - ) - ) + result = LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item(input_item) # Assert expected = { @@ -97,11 +83,7 @@ class TestLiteLLMCompletionResponsesConfig: input_item = {"type": "input_file"} # Execute - result = ( - LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item( - input_item - ) - ) + result = LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item(input_item) # Assert expected = {"type": "file", "file": {}} @@ -120,11 +102,7 @@ class TestLiteLLMCompletionResponsesConfig: } # Execute - result = ( - LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item( - input_item - ) - ) + result = LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item(input_item) # Assert expected = {"type": "file", "file": {"file_id": "file-abc123xyz"}} @@ -134,10 +112,8 @@ class TestLiteLLMCompletionResponsesConfig: def test_transform_input_file_item_to_file_item_with_file_url(self): """file_url should be mapped to file_id for downstream URL handling""" - result = ( - LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item( - {"type": "input_file", "file_url": "https://example.com/doc.pdf"} - ) + result = LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item( + {"type": "input_file", "file_url": "https://example.com/doc.pdf"} ) assert result == { "type": "file", @@ -146,14 +122,12 @@ class TestLiteLLMCompletionResponsesConfig: def test_transform_input_file_item_file_id_takes_precedence_over_file_url(self): """explicit file_id should not be overwritten by file_url""" - result = ( - LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item( - { - "type": "input_file", - "file_id": "file-abc123", - "file_url": "https://example.com/doc.pdf", - } - ) + result = LiteLLMCompletionResponsesConfig._transform_input_file_item_to_file_item( + { + "type": "input_file", + "file_id": "file-abc123", + "file_url": "https://example.com/doc.pdf", + } ) assert result == {"type": "file", "file": {"file_id": "file-abc123"}} @@ -164,11 +138,7 @@ class TestLiteLLMCompletionResponsesConfig: input_item = {"type": "input_image", "image_url": image_url, "detail": "high"} # Execute - result = ( - LiteLLMCompletionResponsesConfig._transform_input_image_item_to_image_item( - input_item - ) - ) + result = LiteLLMCompletionResponsesConfig._transform_input_image_item_to_image_item(input_item) # Assert expected = { @@ -187,11 +157,7 @@ class TestLiteLLMCompletionResponsesConfig: input_item = {"type": "input_image", "image_url": image_url, "detail": "high"} # Execute - result = ( - LiteLLMCompletionResponsesConfig._transform_input_image_item_to_image_item( - input_item - ) - ) + result = LiteLLMCompletionResponsesConfig._transform_input_image_item_to_image_item(input_item) # Assert expected = { @@ -210,11 +176,7 @@ class TestLiteLLMCompletionResponsesConfig: input_item = {"type": "input_image", "image_url": image_url} # Execute - result = ( - LiteLLMCompletionResponsesConfig._transform_input_image_item_to_image_item( - input_item - ) - ) + result = LiteLLMCompletionResponsesConfig._transform_input_image_item_to_image_item(input_item) # Assert expected = { @@ -232,11 +194,7 @@ class TestLiteLLMCompletionResponsesConfig: input_item = {"type": "input_image"} # Execute - result = ( - LiteLLMCompletionResponsesConfig._transform_input_image_item_to_image_item( - input_item - ) - ) + result = LiteLLMCompletionResponsesConfig._transform_input_image_item_to_image_item(input_item) # Assert expected = {"type": "image_url", "image_url": {"url": "", "detail": "auto"}} @@ -256,11 +214,7 @@ class TestLiteLLMCompletionResponsesConfig: } # Execute - result = ( - LiteLLMCompletionResponsesConfig._transform_input_image_item_to_image_item( - input_item - ) - ) + result = LiteLLMCompletionResponsesConfig._transform_input_image_item_to_image_item(input_item) # Assert expected = { @@ -296,28 +250,26 @@ class TestLiteLLMCompletionResponsesConfig: ) # Execute - responses_api_response = LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( - request_input="What is the meaning of life?", - responses_api_request={}, - chat_completion_response=chat_completion_response, + responses_api_response = ( + LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( + request_input="What is the meaning of life?", + responses_api_request={}, + chat_completion_response=chat_completion_response, + ) ) # Assert assert hasattr(responses_api_response, "output") assert len(responses_api_response.output) >= 2 - reasoning_items = [ - item for item in responses_api_response.output if item.type == "reasoning" - ] + reasoning_items = [item for item in responses_api_response.output if item.type == "reasoning"] assert len(reasoning_items) == 1, "Should have exactly one reasoning item" reasoning_item = reasoning_items[0] # Note: ID auto-generation was disabled, so reasoning items may not have IDs # Only assert ID format if an ID is present if hasattr(reasoning_item, "id") and reasoning_item.id: - assert reasoning_item.id.startswith( - "rs_" - ), f"Expected ID to start with 'rs_', got: {reasoning_item.id}" + assert reasoning_item.id.startswith("rs_"), f"Expected ID to start with 'rs_', got: {reasoning_item.id}" assert reasoning_item.status == "completed" assert reasoning_item.role == "assistant" assert len(reasoning_item.content) == 1 @@ -325,9 +277,7 @@ class TestLiteLLMCompletionResponsesConfig: assert "step by step" in reasoning_item.content[0].text assert "42" in reasoning_item.content[0].text - message_items = [ - item for item in responses_api_response.output if item.type == "message" - ] + message_items = [item for item in responses_api_response.output if item.type == "message"] assert len(message_items) == 1, "Should have exactly one message item" message_item = message_items[0] @@ -354,21 +304,19 @@ class TestLiteLLMCompletionResponsesConfig: ) # Execute - responses_api_response = LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( - request_input="A simple question?", - responses_api_request={}, - chat_completion_response=chat_completion_response, + responses_api_response = ( + LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( + request_input="A simple question?", + responses_api_request={}, + chat_completion_response=chat_completion_response, + ) ) # Assert - reasoning_items = [ - item for item in responses_api_response.output if item.type == "reasoning" - ] + reasoning_items = [item for item in responses_api_response.output if item.type == "reasoning"] assert len(reasoning_items) == 0, "Should have no reasoning items" - message_items = [ - item for item in responses_api_response.output if item.type == "message" - ] + message_items = [item for item in responses_api_response.output if item.type == "message"] assert len(message_items) == 1, "Should have exactly one message item" assert message_items[0].content[0].text == "Just a regular answer." assert responses_api_response.object == "response" @@ -404,22 +352,20 @@ class TestLiteLLMCompletionResponsesConfig: ) # Execute - responses_api_response = LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( - request_input="A question with multiple answers?", - responses_api_request={}, - chat_completion_response=chat_completion_response, + responses_api_response = ( + LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( + request_input="A question with multiple answers?", + responses_api_request={}, + chat_completion_response=chat_completion_response, + ) ) # Assert - reasoning_items = [ - item for item in responses_api_response.output if item.type == "reasoning" - ] + reasoning_items = [item for item in responses_api_response.output if item.type == "reasoning"] assert len(reasoning_items) == 1, "Should have exactly one reasoning item" assert reasoning_items[0].content[0].text == "First reasoning process." - message_items = [ - item for item in responses_api_response.output if item.type == "message" - ] + message_items = [item for item in responses_api_response.output if item.type == "message"] assert len(message_items) == 2, "Should have two message items" def test_transform_chat_completion_response_status_with_stop(self): @@ -446,10 +392,12 @@ class TestLiteLLMCompletionResponsesConfig: ], ) - responses_api_response = LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( - request_input="this is a test", - responses_api_request={}, - chat_completion_response=chat_completion_response, + responses_api_response = ( + LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( + request_input="this is a test", + responses_api_request={}, + chat_completion_response=chat_completion_response, + ) ) assert responses_api_response.status == "completed" @@ -485,15 +433,15 @@ class TestLiteLLMCompletionResponsesConfig: ], ) - responses_api_response = LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( - request_input="this is a test", - responses_api_request={}, - chat_completion_response=chat_completion_response, + responses_api_response = ( + LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( + request_input="this is a test", + responses_api_request={}, + chat_completion_response=chat_completion_response, + ) ) - message_items = [ - item for item in responses_api_response.output if item.type == "message" - ] + message_items = [item for item in responses_api_response.output if item.type == "message"] assert len(message_items) > 0 for item in message_items: @@ -528,10 +476,12 @@ class TestLiteLLMCompletionResponsesConfig: ], ) - responses_api_response = LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( - request_input="this is a test", - responses_api_request={}, - chat_completion_response=chat_completion_response, + responses_api_response = ( + LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( + request_input="this is a test", + responses_api_request={}, + chat_completion_response=chat_completion_response, + ) ) assert responses_api_response.status == "incomplete" @@ -563,10 +513,12 @@ class TestLiteLLMCompletionResponsesConfig: } # Execute - responses_api_response = LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( - request_input="Test", - responses_api_request={}, - chat_completion_response=chat_completion_response, + responses_api_response = ( + LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( + request_input="Test", + responses_api_request={}, + chat_completion_response=chat_completion_response, + ) ) # Assert @@ -598,10 +550,12 @@ class TestLiteLLMCompletionResponsesConfig: ) # Execute - responses_api_response = LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( - request_input="Test", - responses_api_request={}, - chat_completion_response=chat_completion_response, + responses_api_response = ( + LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response( + request_input="Test", + responses_api_request={}, + chat_completion_response=chat_completion_response, + ) ) # Assert - should default to empty dict @@ -630,26 +584,14 @@ class TestFunctionCallTransformation: regular_message = {"type": "message", "role": "user", "content": "Hello"} # Test function_call detection - assert LiteLLMCompletionResponsesConfig._is_input_item_function_call( - function_call_item - ) - assert not LiteLLMCompletionResponsesConfig._is_input_item_function_call( - function_call_output_item - ) - assert not LiteLLMCompletionResponsesConfig._is_input_item_function_call( - regular_message - ) + assert LiteLLMCompletionResponsesConfig._is_input_item_function_call(function_call_item) + assert not LiteLLMCompletionResponsesConfig._is_input_item_function_call(function_call_output_item) + assert not LiteLLMCompletionResponsesConfig._is_input_item_function_call(regular_message) # Test function_call_output detection (should still work) - assert LiteLLMCompletionResponsesConfig._is_input_item_tool_call_output( - function_call_output_item - ) - assert not LiteLLMCompletionResponsesConfig._is_input_item_tool_call_output( - function_call_item - ) - assert not LiteLLMCompletionResponsesConfig._is_input_item_tool_call_output( - regular_message - ) + assert LiteLLMCompletionResponsesConfig._is_input_item_tool_call_output(function_call_output_item) + assert not LiteLLMCompletionResponsesConfig._is_input_item_tool_call_output(function_call_item) + assert not LiteLLMCompletionResponsesConfig._is_input_item_tool_call_output(regular_message) def test_function_call_transformation(self): """Test that function_call items are correctly transformed to assistant messages with tool calls""" @@ -733,9 +675,7 @@ class TestFunctionCallTransformation: tool_msg = messages[2] assert tool_msg.get("role") == "tool" assert tool_msg.get("content") == "Rainy" - assert ( - tool_msg.get("tool_call_id") == "call_1fe70e2a-a596-45ef-b72c-9b8567c460e5" - ) + assert tool_msg.get("tool_call_id") == "call_1fe70e2a-a596-45ef-b72c-9b8567c460e5" def test_complete_request_transformation_with_function_calls(self): """Test the complete request transformation that would be used by the responses API""" @@ -940,9 +880,7 @@ class TestToolChoiceTransformation: Test that {"type": "tool"} is transformed to "required". This fixes the Anthropic error: "tool_choice.tool.name: Field required" """ - result = LiteLLMCompletionResponsesConfig._transform_tool_choice( - {"type": "tool"} - ) + result = LiteLLMCompletionResponsesConfig._transform_tool_choice({"type": "tool"}) assert result == "required" def test_transform_tool_choice_preserves_function_with_name(self): @@ -954,23 +892,17 @@ class TestToolChoiceTransformation: def test_transform_tool_choice_responses_flat_function_name(self): """Responses-API forced-function with a top-level name maps to the nested Chat Completions shape instead of degrading to required and dropping the name""" - result = LiteLLMCompletionResponsesConfig._transform_tool_choice( - {"type": "function", "name": "get_weather"} - ) + result = LiteLLMCompletionResponsesConfig._transform_tool_choice({"type": "function", "name": "get_weather"}) assert result == {"type": "function", "function": {"name": "get_weather"}} def test_transform_tool_choice_function_without_name_falls_back_to_required(self): """A function-type dict with no name still falls back to required""" - result = LiteLLMCompletionResponsesConfig._transform_tool_choice( - {"type": "function"} - ) + result = LiteLLMCompletionResponsesConfig._transform_tool_choice({"type": "function"}) assert result == "required" def test_transform_tool_choice_function_empty_name_falls_back_to_required(self): """An empty top-level name is falsy and must not produce an empty function name""" - result = LiteLLMCompletionResponsesConfig._transform_tool_choice( - {"type": "function", "name": ""} - ) + result = LiteLLMCompletionResponsesConfig._transform_tool_choice({"type": "function", "name": ""}) assert result == "required" @@ -982,20 +914,12 @@ class TestContentTypeTransformation: Test that 'tool_result' content type is transformed to 'text'. This fixes: Invalid user message - content type 'tool_result' not valid. """ - result = ( - LiteLLMCompletionResponsesConfig._get_chat_completion_request_content_type( - "tool_result" - ) - ) + result = LiteLLMCompletionResponsesConfig._get_chat_completion_request_content_type("tool_result") assert result == "text" def test_input_text_content_type_transformed_to_text(self): """Test that 'input_text' content type is transformed to 'text'""" - result = ( - LiteLLMCompletionResponsesConfig._get_chat_completion_request_content_type( - "input_text" - ) - ) + result = LiteLLMCompletionResponsesConfig._get_chat_completion_request_content_type("input_text") assert result == "text" def test_none_text_blocks_filtered_out(self): @@ -1009,9 +933,7 @@ class TestContentTypeTransformation: {"type": "text", "text": None}, # Should be filtered out {"type": "text", "text": "another valid"}, ] - result = LiteLLMCompletionResponsesConfig._transform_responses_api_content_to_chat_completion_content( - content - ) + result = LiteLLMCompletionResponsesConfig._transform_responses_api_content_to_chat_completion_content(content) assert len(result) == 2 assert result[0]["text"] == "valid text" assert result[1]["text"] == "another valid" @@ -1033,9 +955,7 @@ class TestToolTransformation: ( result_tools, web_search_options, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 1 @@ -1057,9 +977,7 @@ class TestToolTransformation: ( result_tools, web_search_options, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 1 @@ -1087,9 +1005,7 @@ class TestToolTransformation: ( result_tools, web_search_options, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert - computer_use has no Chat Completions equivalent, so it is dropped assert len(result_tools) == 0 @@ -1114,9 +1030,7 @@ class TestToolTransformation: ( result_tools, web_search_options, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert - custom tool is converted to a function tool assert len(result_tools) == 1 @@ -1141,9 +1055,7 @@ class TestToolTransformation: ( result_tools, web_search_options, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 1 @@ -1167,9 +1079,7 @@ class TestToolTransformation: ( result_tools, _, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 1 @@ -1190,9 +1100,7 @@ class TestToolTransformation: ( result_tools, _, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 1 @@ -1211,9 +1119,7 @@ class TestToolTransformation: tools = [custom_tool] with pytest.raises(ValueError, match="allowed_callers must be a list of strings"): - LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) def test_transform_web_search_tools_to_web_search_options(self): """Test that web_search tools are converted to web_search_options""" @@ -1229,9 +1135,7 @@ class TestToolTransformation: ( result_tools, web_search_options, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 0 # Web search is not added to tools @@ -1262,9 +1166,7 @@ class TestToolTransformation: ( result_tools, web_search_options, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 1 @@ -1294,9 +1196,7 @@ class TestToolTransformation: ( result_tools, _, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 1 @@ -1322,9 +1222,7 @@ class TestToolTransformation: ( result_tools, _, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 1 @@ -1350,9 +1248,7 @@ class TestToolTransformation: ( result_tools, _, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 1 @@ -1376,9 +1272,7 @@ class TestToolTransformation: ( result_tools, _, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 2 @@ -1410,14 +1304,10 @@ class TestToolTransformation: ( result_tools, web_search_options, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert - assert ( - len(result_tools) == 3 - ) # function, mcp, vertex (web_search becomes options) + assert len(result_tools) == 3 # function, mcp, vertex (web_search becomes options) assert web_search_options is not None # Check function tool @@ -1447,9 +1337,7 @@ class TestToolTransformation: ( result_tools, _, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 1 @@ -1472,9 +1360,7 @@ class TestToolTransformation: ( result_tools, _, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 1 @@ -1495,9 +1381,7 @@ class TestToolTransformation: ( result_tools, _, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 1 @@ -1519,19 +1403,14 @@ class TestToolTransformation: ( result_tools, _, - ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools( - tools=tools - ) + ) = LiteLLMCompletionResponsesConfig.transform_responses_api_tools_to_chat_completion_tools(tools=tools) # Assert assert len(result_tools) == 1 result_tool = result_tools[0] assert result_tool["function"]["parameters"]["type"] == "object" assert "properties" in result_tool["function"]["parameters"] - assert ( - result_tool["function"]["parameters"]["properties"]["arg"]["type"] - == "string" - ) + assert result_tool["function"]["parameters"]["properties"]["arg"]["type"] == "string" def test_bedrock_anthropic_drops_derived_web_search_options(self): """ @@ -1657,11 +1536,7 @@ class TestToolTransformation: assert "web_search_options" not in result bedrock_tool_blocks = _bedrock_tools_pt(tools=result["tools"], model=model) - names = [ - block["toolSpec"]["name"] - for block in bedrock_tool_blocks - if "toolSpec" in block - ] + names = [block["toolSpec"]["name"] for block in bedrock_tool_blocks if "toolSpec" in block] assert names == ["noop"] assert not any(name.startswith("litellm_unnamed_tool_") for name in names) @@ -1940,9 +1815,7 @@ class TestUsageTransformation: Choices( finish_reason="stop", index=0, - message=Message( - content="Here is the generated image.", role="assistant" - ), + message=Message(content="Here is the generated image.", role="assistant"), ) ], ) @@ -2166,9 +2039,7 @@ class TestStreamingIDConsistency: # Verify the cached ID is set and matches assert iterator._cached_item_id is not None, "Iterator should cache the item_id" assert iterator._cached_item_id == item_id_1, "Cached ID should match event IDs" - assert ( - iterator._cached_item_id == "chatcmpl-first-id" - ), "Should use the first chunk's ID" + assert iterator._cached_item_id == "chatcmpl-first-id", "Should use the first chunk's ID" def test_streaming_iterator_initial_events_use_cached_id(self): """ @@ -2258,9 +2129,7 @@ class TestStreamingIDConsistency: # Create done events text_done_event = iterator.create_output_text_done_event(complete_response) - content_done_event = iterator.create_output_content_part_done_event( - complete_response - ) + content_done_event = iterator.create_output_content_part_done_event(complete_response) item_done_event = iterator.create_output_item_done_event(complete_response) # Extract IDs @@ -2320,27 +2189,19 @@ class TestStreamingIDConsistency: input=input_items ) - roles = [ - m.get("role") if isinstance(m, dict) else getattr(m, "role", None) - for m in messages - ] + roles = [m.get("role") if isinstance(m, dict) else getattr(m, "role", None) for m in messages] # Must not have two consecutive assistant messages for i in range(len(roles) - 1): - assert not ( - roles[i] == "assistant" and roles[i + 1] == "assistant" - ), f"Consecutive assistant messages at indices {i} and {i+1}: {roles}" + assert not (roles[i] == "assistant" and roles[i + 1] == "assistant"), ( + f"Consecutive assistant messages at indices {i} and {i + 1}: {roles}" + ) # The single assistant message must contain BOTH tool_calls assistant_messages = [ - m - for m in messages - if (m.get("role") if isinstance(m, dict) else getattr(m, "role", None)) - == "assistant" + m for m in messages if (m.get("role") if isinstance(m, dict) else getattr(m, "role", None)) == "assistant" ] - assert ( - len(assistant_messages) == 1 - ), f"Expected 1 assistant message, got {len(assistant_messages)}" + assert len(assistant_messages) == 1, f"Expected 1 assistant message, got {len(assistant_messages)}" assistant_msg = assistant_messages[0] tool_calls = ( @@ -2348,27 +2209,19 @@ class TestStreamingIDConsistency: if isinstance(assistant_msg, dict) else getattr(assistant_msg, "tool_calls", None) ) - assert ( - tool_calls is not None and len(tool_calls) == 2 - ), f"Expected 2 tool_calls in the merged assistant message, got: {tool_calls}" + assert tool_calls is not None and len(tool_calls) == 2, ( + f"Expected 2 tool_calls in the merged assistant message, got: {tool_calls}" + ) - call_ids = [ - (tc.get("id") if isinstance(tc, dict) else getattr(tc, "id", None)) - for tc in tool_calls - ] + call_ids = [(tc.get("id") if isinstance(tc, dict) else getattr(tc, "id", None)) for tc in tool_calls] assert "toolu_01" in call_ids, f"toolu_01 missing from tool_calls: {call_ids}" assert "toolu_02" in call_ids, f"toolu_02 missing from tool_calls: {call_ids}" # Both tool messages must be present tool_messages = [ - m - for m in messages - if (m.get("role") if isinstance(m, dict) else getattr(m, "role", None)) - == "tool" + m for m in messages if (m.get("role") if isinstance(m, dict) else getattr(m, "role", None)) == "tool" ] - assert ( - len(tool_messages) == 2 - ), f"Expected 2 tool messages, got {len(tool_messages)}" + assert len(tool_messages) == 2, f"Expected 2 tool messages, got {len(tool_messages)}" def test_single_tool_call_still_works_after_merge_fix(self): """ @@ -2390,20 +2243,14 @@ class TestStreamingIDConsistency: input=input_items ) - roles = [ - m.get("role") if isinstance(m, dict) else getattr(m, "role", None) - for m in messages - ] + roles = [m.get("role") if isinstance(m, dict) else getattr(m, "role", None) for m in messages] assert "user" in roles assert "assistant" in roles assert "tool" in roles assistant_messages = [ - m - for m in messages - if (m.get("role") if isinstance(m, dict) else getattr(m, "role", None)) - == "assistant" + m for m in messages if (m.get("role") if isinstance(m, dict) else getattr(m, "role", None)) == "assistant" ] assert len(assistant_messages) == 1 @@ -2576,9 +2423,7 @@ class TestEnsureOutputItemContentPartAdded: LiteLLMCompletionStreamingIterator, ) - iterator = LiteLLMCompletionStreamingIterator.__new__( - LiteLLMCompletionStreamingIterator - ) + iterator = LiteLLMCompletionStreamingIterator.__new__(LiteLLMCompletionStreamingIterator) iterator.sent_output_item_added_event = False iterator.sent_content_part_added_event = False iterator._sequence_number = 0 @@ -2671,9 +2516,7 @@ class TestEnsureOutputItemContentPartAdded: usage=Usage(prompt_tokens=10, completion_tokens=1, total_tokens=11), ) - completed_event = iterator._emit_response_completed_event( - litellm_model_response - ) + completed_event = iterator._emit_response_completed_event(litellm_model_response) assert completed_event is not None assert completed_event.response.status == "incomplete" @@ -2716,9 +2559,7 @@ class TestCacheControlPreservation: "cache_control": {"type": "ephemeral"}, } ] - result = LiteLLMCompletionResponsesConfig._transform_responses_api_content_to_chat_completion_content( - content - ) + result = LiteLLMCompletionResponsesConfig._transform_responses_api_content_to_chat_completion_content(content) assert isinstance(result, list) assert len(result) == 1 assert result[0]["cache_control"] == {"type": "ephemeral"} @@ -2726,9 +2567,7 @@ class TestCacheControlPreservation: def test_content_without_cache_control_unaffected(self): """Content blocks that don't have cache_control should be unaffected.""" content = [{"type": "text", "text": "hello"}] - result = LiteLLMCompletionResponsesConfig._transform_responses_api_content_to_chat_completion_content( - content - ) + result = LiteLLMCompletionResponsesConfig._transform_responses_api_content_to_chat_completion_content(content) assert isinstance(result, list) assert len(result) == 1 assert "cache_control" not in result[0] @@ -2750,9 +2589,7 @@ class TestCacheControlPreservation: ) assert len(messages) == 1 msg_content = ( - messages[0].get("content") - if isinstance(messages[0], dict) - else getattr(messages[0], "content", None) + messages[0].get("content") if isinstance(messages[0], dict) else getattr(messages[0], "content", None) ) assert isinstance(msg_content, list) assert msg_content[0]["cache_control"] == {"type": "ephemeral"} @@ -2765,9 +2602,7 @@ class TestCacheControlPreservation: "cache_control": {"type": "ephemeral"}, } ] - result = LiteLLMCompletionResponsesConfig._transform_responses_api_content_to_chat_completion_content( - content - ) + result = LiteLLMCompletionResponsesConfig._transform_responses_api_content_to_chat_completion_content(content) assert isinstance(result, list) assert len(result) == 1 assert result[0]["cache_control"] == {"type": "ephemeral"} @@ -2780,13 +2615,121 @@ class TestCacheControlPreservation: "cache_control": {"type": "ephemeral"}, } ] - result = LiteLLMCompletionResponsesConfig._transform_responses_api_content_to_chat_completion_content( - content - ) + result = LiteLLMCompletionResponsesConfig._transform_responses_api_content_to_chat_completion_content(content) assert isinstance(result, list) assert len(result) == 1 assert result[0]["cache_control"] == {"type": "ephemeral"} + def test_cache_control_preserved_for_object_input_item(self): + """Test that cache_control is preserved when input_item is a custom object / model.""" + + class MockInputItem: + def __init__(self): + self.role = "user" + self.content = "hello" + self.cache_control = {"type": "ephemeral"} + + input_item = MockInputItem() + messages = LiteLLMCompletionResponsesConfig._transform_responses_api_input_item_to_chat_completion_message( + input_item + ) + assert len(messages) == 1 + assert messages[0].get("cache_control") == {"type": "ephemeral"} + + def test_cache_control_preserved_for_object_content_item(self): + """Test that cache_control is preserved when content items are custom objects.""" + + class MockContentBlock: + def __init__(self): + self.type = "text" + self.text = "hello" + self.cache_control = {"type": "ephemeral"} + + class MockPydanticV2Block: + def __init__(self): + self.type = "text" + self.text = "hello v2" + self.cache_control = {"type": "ephemeral"} + + def model_dump(self): + return {"type": self.type, "text": self.text, "cache_control": self.cache_control} + + class MockPydanticV1Block: + def __init__(self): + self.type = "text" + self.text = "hello v1" + self.cache_control = {"type": "ephemeral"} + + def dict(self): + return {"type": self.type, "text": self.text, "cache_control": self.cache_control} + + content = [MockContentBlock(), MockPydanticV2Block(), MockPydanticV1Block()] + result = LiteLLMCompletionResponsesConfig._transform_responses_api_content_to_chat_completion_content(content) + assert isinstance(result, list) + assert len(result) == 3 + assert result[0]["cache_control"] == {"type": "ephemeral"} + assert result[1]["cache_control"] == {"type": "ephemeral"} + assert result[2]["cache_control"] == {"type": "ephemeral"} + assert result[1]["text"] == "hello v2" + assert result[2]["text"] == "hello v1" + + def test_is_cached_message_for_object_message_and_content_item(self): + """Test is_cached_message on custom objects / models.""" + from litellm.utils import is_cached_message + + # Test message level cache_control object + class MockCacheControl: + def __init__(self): + self.type = "ephemeral" + + class MockMessageLevelObj: + def __init__(self): + self.role = "system" + self.content = "hello" + self.cache_control = MockCacheControl() + + msg = MockMessageLevelObj() + assert is_cached_message(msg) is True + + # Test content level cache_control object + class MockContentItem: + def __init__(self): + self.type = "text" + self.text = "hello" + self.cache_control = MockCacheControl() + + class MockContentLevelObj: + def __init__(self): + self.role = "system" + self.content = [MockContentItem()] + + msg = MockContentLevelObj() + assert is_cached_message(msg) is True + + def test_extract_ttl_from_cached_messages_for_object_models(self): + """Test extract_ttl_from_cached_messages with object-based messages and content items.""" + from litellm.llms.vertex_ai.context_caching.transformation import extract_ttl_from_cached_messages + + class MockCacheControl: + def __init__(self): + self.type = "ephemeral" + self.ttl = "3600s" + + class MockContentItem: + def __init__(self): + self.type = "text" + self.text = "hello" + self.cache_control = MockCacheControl() + + class MockMessageObj: + def __init__(self): + self.role = "system" + self.content = [MockContentItem()] + + messages = [MockMessageObj()] + ttl = extract_ttl_from_cached_messages(messages) + assert ttl == "3600s" + def test_function_call_tool_id_falls_back_to_unique_id_for_degenerate_call_id(): """Bedrock Mantle returns a non-unique, index-based ``call_id`` (``call_0`` that @@ -2798,16 +2741,10 @@ def test_function_call_tool_id_falls_back_to_unique_id_for_degenerate_call_id(): for the bedrock-mantle gpt-5.5 non-streaming path.""" from types import SimpleNamespace - convert = ( - LiteLLMCompletionResponsesConfig.convert_response_function_tool_call_to_chat_completion_tool_call - ) + convert = LiteLLMCompletionResponsesConfig.convert_response_function_tool_call_to_chat_completion_tool_call - mantle = SimpleNamespace( - id="fc_unique_abc123", call_id="call_0", name="get_weather", arguments="{}" - ) + mantle = SimpleNamespace(id="fc_unique_abc123", call_id="call_0", name="get_weather", arguments="{}") assert convert(mantle)["id"] == "fc_unique_abc123" - openai = SimpleNamespace( - id="fc_2", call_id="call_tokyo", name="get_weather", arguments="{}" - ) + openai = SimpleNamespace(id="fc_2", call_id="call_tokyo", name="get_weather", arguments="{}") assert convert(openai)["id"] == "call_tokyo"