The agent engineering platform.
Python agentsaiai-agentsanthropicchatgptdeepagentsenterpriseframeworkgeminigenerative-ailangchainlanggraphllmmultiagentopen-sourceopenaipydanticpythonragtypescript
fix(core): fix string content when streaming output_version="v1" (#33261)
ccurme7404338786b4
diff --git a/libs/core/langchain_core/language_models/chat_models.py b/libs/core/langchain_core/language_models/chat_models.pyindex a4acd7683d..c499492502 100644--- a/libs/core/langchain_core/language_models/chat_models.py+++ b/libs/core/langchain_core/language_models/chat_models.py@@ -43,6 +43,7 @@ from langchain_core.messages import (is_data_content_block,message_chunk_to_message,)+from langchain_core.messages import content as typesfrom langchain_core.messages.block_translators.openai import (convert_to_openai_image_block,)@@ -533,6 +534,8 @@ class BaseChatModel(BaseLanguageModel[AIMessage], ABC):input_messages = _normalize_messages(messages)run_id = "-".join((LC_ID_PREFIX, str(run_manager.run_id)))yielded = False+ index = -1+ index_type = ""for chunk in self._stream(input_messages, stop=stop, **kwargs):if chunk.message.id is None:chunk.message.id = run_id@@ -542,6 +545,14 @@ class BaseChatModel(BaseLanguageModel[AIMessage], ABC):chunk.message = _update_message_content_to_blocks(chunk.message, "v1")+ for block in cast(+ "list[types.ContentBlock]", chunk.message.content+ ):+ if block["type"] != index_type:+ index_type = block["type"]+ index = index + 1+ if "index" not in block:+ block["index"] = indexrun_manager.on_llm_new_token(cast("str", chunk.message.content), chunk=chunk)@@ -651,6 +662,8 @@ class BaseChatModel(BaseLanguageModel[AIMessage], ABC):input_messages = _normalize_messages(messages)run_id = "-".join((LC_ID_PREFIX, str(run_manager.run_id)))yielded = False+ index = -1+ index_type = ""async for chunk in self._astream(input_messages,stop=stop,@@ -664,6 +677,14 @@ class BaseChatModel(BaseLanguageModel[AIMessage], ABC):chunk.message = _update_message_content_to_blocks(chunk.message, "v1")+ for block in cast(+ "list[types.ContentBlock]", chunk.message.content+ ):+ if block["type"] != index_type:+ index_type = block["type"]+ index = index + 1+ if "index" not in block:+ block["index"] = indexawait run_manager.on_llm_new_token(cast("str", chunk.message.content), chunk=chunk)@@ -1145,6 +1166,8 @@ class BaseChatModel(BaseLanguageModel[AIMessage], ABC):f"{LC_ID_PREFIX}-{run_manager.run_id}" if run_manager else None)yielded = False+ index = -1+ index_type = ""for chunk in self._stream(messages, stop=stop, **kwargs):chunk.message.response_metadata = _gen_info_and_msg_metadata(chunk)if self.output_version == "v1":@@ -1152,6 +1175,14 @@ class BaseChatModel(BaseLanguageModel[AIMessage], ABC):chunk.message = _update_message_content_to_blocks(chunk.message, "v1")+ for block in cast(+ "list[types.ContentBlock]", chunk.message.content+ ):+ if block["type"] != index_type:+ index_type = block["type"]+ index = index + 1+ if "index" not in block:+ block["index"] = indexif run_manager:if chunk.message.id is None:chunk.message.id = run_id@@ -1253,6 +1284,8 @@ class BaseChatModel(BaseLanguageModel[AIMessage], ABC):f"{LC_ID_PREFIX}-{run_manager.run_id}" if run_manager else None)yielded = False+ index = -1+ index_type = ""async for chunk in self._astream(messages, stop=stop, **kwargs):chunk.message.response_metadata = _gen_info_and_msg_metadata(chunk)if self.output_version == "v1":@@ -1260,6 +1293,14 @@ class BaseChatModel(BaseLanguageModel[AIMessage], ABC):chunk.message = _update_message_content_to_blocks(chunk.message, "v1")+ for block in cast(+ "list[types.ContentBlock]", chunk.message.content+ ):+ if block["type"] != index_type:+ index_type = block["type"]+ index = index + 1+ if "index" not in block:+ block["index"] = indexif run_manager:if chunk.message.id is None:chunk.message.id = run_iddiff --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 9820d4ba2a..cdcc0fa637 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@@ -8,7 +8,10 @@ from typing import TYPE_CHECKING, Any, Literalimport pytestfrom typing_extensions import override-from langchain_core.callbacks import CallbackManagerForLLMRun+from langchain_core.callbacks import (+ AsyncCallbackManagerForLLMRun,+ CallbackManagerForLLMRun,+)from langchain_core.language_models import (BaseChatModel,FakeListChatModel,@@ -23,7 +26,6 @@ from langchain_core.messages import (AIMessage,AIMessageChunk,BaseMessage,- BaseMessageChunk,HumanMessage,SystemMessage,)@@ -907,6 +909,56 @@ async def test_output_version_ainvoke(monkeypatch: Any) -> None:assert response.response_metadata["output_version"] == "v1"+class _AnotherFakeChatModel(BaseChatModel):+ responses: Iterator[AIMessage]+ """Responses for _generate."""++ chunks: Iterator[AIMessageChunk]+ """Responses for _stream."""++ @property+ def _llm_type(self) -> str:+ return "another-fake-chat-model"++ def _generate(+ self,+ messages: list[BaseMessage], # noqa: ARG002+ stop: list[str] | None = None, # noqa: ARG002+ run_manager: CallbackManagerForLLMRun | None = None, # noqa: ARG002+ **kwargs: Any, # noqa: ARG002+ ) -> ChatResult:+ return ChatResult(generations=[ChatGeneration(message=next(self.responses))])++ async def _agenerate(+ self,+ messages: list[BaseMessage], # noqa: ARG002+ stop: list[str] | None = None, # noqa: ARG002+ run_manager: AsyncCallbackManagerForLLMRun | None = None, # noqa: ARG002+ **kwargs: Any, # noqa: ARG002+ ) -> ChatResult:+ return ChatResult(generations=[ChatGeneration(message=next(self.responses))])++ def _stream(+ self,+ messages: list[BaseMessage], # noqa: ARG002+ stop: list[str] | None = None, # noqa: ARG002+ run_manager: CallbackManagerForLLMRun | None = None, # noqa: ARG002+ **kwargs: Any, # noqa: ARG002+ ) -> Iterator[ChatGenerationChunk]:+ for chunk in self.chunks:+ yield ChatGenerationChunk(message=chunk)++ async def _astream(+ self,+ messages: list[BaseMessage], # noqa: ARG002+ stop: list[str] | None = None, # noqa: ARG002+ run_manager: AsyncCallbackManagerForLLMRun | None = None, # noqa: ARG002+ **kwargs: Any, # noqa: ARG002+ ) -> AsyncIterator[ChatGenerationChunk]:+ for chunk in self.chunks:+ yield ChatGenerationChunk(message=chunk)++def test_output_version_stream(monkeypatch: Any) -> None:messages = [AIMessage("foo bar")]@@ -923,7 +975,7 @@ def test_output_version_stream(monkeypatch: Any) -> None:# v1llm = GenericFakeChatModel(messages=iter(messages), output_version="v1")- full_v1: BaseMessageChunk | None = None+ full_v1: AIMessageChunk | None = Nonefor chunk in llm.stream("hello"):assert isinstance(chunk, AIMessageChunk)assert isinstance(chunk.content, list)@@ -936,6 +988,58 @@ def test_output_version_stream(monkeypatch: Any) -> None:assert isinstance(full_v1, AIMessageChunk)assert full_v1.response_metadata["output_version"] == "v1"+ assert full_v1.content == [{"type": "text", "text": "foo bar", "index": 0}]++ # Test text blocks+ llm_with_rich_content = _AnotherFakeChatModel(+ responses=iter([]),+ chunks=iter(+ [+ AIMessageChunk(content="foo "),+ AIMessageChunk(content="bar"),+ ]+ ),+ output_version="v1",+ )+ full_v1 = None+ for chunk in llm_with_rich_content.stream("hello"):+ full_v1 = chunk if full_v1 is None else full_v1 + chunk+ assert isinstance(full_v1, AIMessageChunk)+ assert full_v1.content_blocks == [{"type": "text", "text": "foo bar", "index": 0}]++ # Test content blocks of different types+ chunks = [+ AIMessageChunk(content="", additional_kwargs={"reasoning_content": "<rea"}),+ AIMessageChunk(content="", additional_kwargs={"reasoning_content": "soning>"}),+ AIMessageChunk(content="<some "),+ AIMessageChunk(content="text>"),+ ]+ llm_with_rich_content = _AnotherFakeChatModel(+ responses=iter([]),+ chunks=iter(chunks),+ output_version="v1",+ )+ full_v1 = None+ for chunk in llm_with_rich_content.stream("hello"):+ full_v1 = chunk if full_v1 is None else full_v1 + chunk+ assert isinstance(full_v1, AIMessageChunk)+ assert full_v1.content_blocks == [+ {"type": "reasoning", "reasoning": "<reasoning>", "index": 0},+ {"type": "text", "text": "<some text>", "index": 1},+ ]++ # Test invoke with stream=True+ llm_with_rich_content = _AnotherFakeChatModel(+ responses=iter([]),+ chunks=iter(chunks),+ output_version="v1",+ )+ response_v1 = llm_with_rich_content.invoke("hello", stream=True)+ assert response_v1.content_blocks == [+ {"type": "reasoning", "reasoning": "<reasoning>", "index": 0},+ {"type": "text", "text": "<some text>", "index": 1},+ ]+# v1 from env varmonkeypatch.setenv("LC_OUTPUT_VERSION", "v1")llm = GenericFakeChatModel(messages=iter(messages))@@ -969,7 +1073,7 @@ async def test_output_version_astream(monkeypatch: Any) -> None:# v1llm = GenericFakeChatModel(messages=iter(messages), output_version="v1")- full_v1: BaseMessageChunk | None = None+ full_v1: AIMessageChunk | None = Noneasync for chunk in llm.astream("hello"):assert isinstance(chunk, AIMessageChunk)assert isinstance(chunk.content, list)@@ -982,6 +1086,58 @@ async def test_output_version_astream(monkeypatch: Any) -> None:assert isinstance(full_v1, AIMessageChunk)assert full_v1.response_metadata["output_version"] == "v1"+ assert full_v1.content == [{"type": "text", "text": "foo bar", "index": 0}]++ # Test text blocks+ llm_with_rich_content = _AnotherFakeChatModel(+ responses=iter([]),+ chunks=iter(+ [+ AIMessageChunk(content="foo "),+ AIMessageChunk(content="bar"),+ ]+ ),+ output_version="v1",+ )+ full_v1 = None+ async for chunk in llm_with_rich_content.astream("hello"):+ full_v1 = chunk if full_v1 is None else full_v1 + chunk+ assert isinstance(full_v1, AIMessageChunk)+ assert full_v1.content_blocks == [{"type": "text", "text": "foo bar", "index": 0}]++ # Test content blocks of different types+ chunks = [+ AIMessageChunk(content="", additional_kwargs={"reasoning_content": "<rea"}),+ AIMessageChunk(content="", additional_kwargs={"reasoning_content": "soning>"}),+ AIMessageChunk(content="<some "),+ AIMessageChunk(content="text>"),+ ]+ llm_with_rich_content = _AnotherFakeChatModel(+ responses=iter([]),+ chunks=iter(chunks),+ output_version="v1",+ )+ full_v1 = None+ async for chunk in llm_with_rich_content.astream("hello"):+ full_v1 = chunk if full_v1 is None else full_v1 + chunk+ assert isinstance(full_v1, AIMessageChunk)+ assert full_v1.content_blocks == [+ {"type": "reasoning", "reasoning": "<reasoning>", "index": 0},+ {"type": "text", "text": "<some text>", "index": 1},+ ]++ # Test invoke with stream=True+ llm_with_rich_content = _AnotherFakeChatModel(+ responses=iter([]),+ chunks=iter(chunks),+ output_version="v1",+ )+ response_v1 = await llm_with_rich_content.ainvoke("hello", stream=True)+ assert response_v1.content_blocks == [+ {"type": "reasoning", "reasoning": "<reasoning>", "index": 0},+ {"type": "text", "text": "<some text>", "index": 1},+ ]+# v1 from env varmonkeypatch.setenv("LC_OUTPUT_VERSION", "v1")llm = GenericFakeChatModel(messages=iter(messages))