The agent engineering platform.
Python agentsaiai-agentsanthropicchatgptdeepagentsenterpriseframeworkgeminigenerative-ailangchainlanggraphllmmultiagentopen-sourceopenaipydanticpythonragtypescript
core[patch]: fix edge cases for _is_openai_data_block (#30997)
ccurmef4863f82e24f
diff --git a/libs/core/langchain_core/language_models/_utils.py b/libs/core/langchain_core/language_models/_utils.pyindex bc2285f920..f54d69728d 100644--- a/libs/core/langchain_core/language_models/_utils.py+++ b/libs/core/langchain_core/language_models/_utils.py@@ -1,4 +1,5 @@import re+from collections.abc import Sequencefrom typing import Optionalfrom langchain_core.messages import BaseMessage@@ -7,24 +8,30 @@ from langchain_core.messages import BaseMessagedef _is_openai_data_block(block: dict) -> bool:"""Check if the block contains multimodal data in OpenAI Chat Completions format."""if block.get("type") == "image_url":- url = block.get("image_url", {}).get("url")- if isinstance(url, str) and set(block.keys()) <= {- "type",- "image_url",- "detail",- }:- return True+ if (+ (set(block.keys()) <= {"type", "image_url", "detail"})+ and (image_url := block.get("image_url"))+ and isinstance(image_url, dict)+ ):+ url = image_url.get("url")+ if isinstance(url, str):+ return Trueelif block.get("type") == "file":- data = block.get("file", {}).get("file_data")- if isinstance(data, str):- return True+ if (file := block.get("file")) and isinstance(file, dict):+ file_data = file.get("file_data")+ if isinstance(file_data, str):+ return True- elif block.get("type") == "input_audio":- audio_data = block.get("input_audio", {}).get("data")- audio_format = block.get("input_audio", {}).get("format")- if isinstance(audio_data, str) and isinstance(audio_format, str):- return True+ elif block.get("type") == "input_audio": # noqa: SIM102+ if (input_audio := block.get("input_audio")) and isinstance(input_audio, dict):+ audio_data = input_audio.get("data")+ audio_format = input_audio.get("format")+ if isinstance(audio_data, str) and isinstance(audio_format, str):+ return True++ else:+ return Falsereturn False@@ -98,7 +105,7 @@ def _convert_openai_format_to_data_block(block: dict) -> dict:return block-def _normalize_messages(messages: list[BaseMessage]) -> list[BaseMessage]:+def _normalize_messages(messages: Sequence[BaseMessage]) -> list[BaseMessage]:"""Extend support for message formats.Chat models implement support for images in OpenAI Chat Completions format, as welldiff --git a/libs/core/tests/unit_tests/language_models/chat_models/test_base.py b/libs/core/tests/unit_tests/language_models/chat_models/test_base.pyindex 99c2a829b8..00e7c8e9f2 100644--- a/libs/core/tests/unit_tests/language_models/chat_models/test_base.py+++ b/libs/core/tests/unit_tests/language_models/chat_models/test_base.py@@ -13,6 +13,7 @@ from langchain_core.language_models import (FakeListChatModel,ParrotFakeChatModel,)+from langchain_core.language_models._utils import _normalize_messagesfrom langchain_core.language_models.fake_chat_models import FakeListChatModelErrorfrom langchain_core.messages import (AIMessage,@@ -567,3 +568,31 @@ def test_extend_support_to_openai_multimodal_formats() -> None:"input_audio": {"data": "<base64 data>", "format": "wav"},},]+++def test_normalize_messages_edge_cases() -> None:+ # Test some blocks that should pass through+ messages = [+ HumanMessage(+ content=[+ {+ "type": "file",+ "file": "uri",+ },+ {+ "type": "input_file",+ "file_data": "uri",+ "filename": "file-name",+ },+ {+ "type": "input_audio",+ "input_audio": "uri",+ },+ {+ "type": "input_image",+ "image_url": "uri",+ },+ ]+ )+ ]+ assert messages == _normalize_messages(messages)