tool calls args missing when yield AIMessageChunk

[dragonet5]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

from langchain_core.messages import (
    AIMessage,
    AIMessageChunk,
    BaseMessage,
    BaseMessageChunk,
    ChatMessage,
    ChatMessageChunk,
    FunctionMessage,
    FunctionMessageChunk,
    HumanMessage,
    HumanMessageChunk,
    InvalidToolCall,
    SystemMessage,
    SystemMessageChunk,
    ToolCall,
    ToolMessage,
    ToolMessageChunk,
)


# the previous tool_call_chunk generated a invalid_tool_calls which is as expected.
# invalid_tool_calls=[{'name': 'get_current_weather', 'args': None, 'id': 'call_cvjtfa42c3m2t75bs2jg', 'error': None, 'type': 'invalid_tool_call'}]

# the new tool_call_chunk parsed from the llm response
tool_call_chunks = [{'name': None, 'args': '{"location": "北京"}', 'id': 'call_cvjtbns2c3m5i17c9fr0', 'type': 'function', 'index':0}]

# try to yield the new tool_call_chunk and generate a full tool_call response
yield AIMessageChunk(
    content='',
    # invalid_tool_calls = invalid_tool_calls,
    tool_call_chunks = tool_call_chunks
)

Description

I'm trying to implement a customized chat model supporting tool calls under langchain frame, the batch mode works as expected, but streaming mode doesn't.

The first step: I write a function to parse the llm respons, the first tool_call_chunk as below:

tool_call_chunks = [{'name': 'get_current_weather', 'args': None, 'id': 'call_cvjtbns2c3m5i17c9fr0', 'type': 'function', 'index':0}]

and then yield the AIMessageChunk. chat_model response as expected:

content='' additional_kwargs={} response_metadata={} id='run-ff870083-85d7-46c0-a408-390b0f12f54e' invalid_tool_calls=[{'name': 'get_current_weather', 'args': None, 'id': 'call_cvl610c2c3m5au8rjq80', 'error': None, 'type': 'invalid_tool_call'}] usage_metadata={'input_tokens': 7, 'output_tokens': 0, 'total_tokens': 7} tool_call_chunks=[{'name': 'get_current_weather', 'args': None, 'id': 'call_cvl610c2c3m5au8rjq80', 'index': None, 'type': 'tool_call_chunk'}] finish_reason=None

The second step I yield the second tool_call_chunk:

[{'name': None, 'args': '{"location":"北京"}', 'id': 'call_cvl610c2c3m5au8rjq80', 'type': 'function', 'index': 0}]

and yield the AIMessageChunk, the tool_call info generated, but the name parameter in args was missing, I don't know what's the problem.

content='' additional_kwargs={} response_metadata={} id='run-ff870083-85d7-46c0-a408-390b0f12f54e' tool_calls=[{'name': '', 'args': {'location': '北京'}, 'id': 'call_cvl610c2c3m5au8rjq80', 'type': 'tool_call'}] usage_metadata={'input_tokens': 7, 'output_tokens': 0, 'total_tokens': 7} tool_call_chunks=[{'name': None, 'args': '{"location":"北京"}', 'id': 'call_cvl610c2c3m5au8rjq80', 'index': 0, 'type': 'tool_call_chunk'}] reasoning_content='' reasoning_flag=0 finish_reason=None

System Info

platform - win10 64bit

conda list | findstr "langchain"

langchain 0.3.21 pymin39_hff2d567_0 conda-forge
langchain-community 0.3.20 pymin39_hff2d567_0 conda-forge
langchain-core 0.3.46 pyhd8ed1ab_0 conda-forge
langchain-openai 0.1.23 pypi_0 pypi
langchain-text-splitters 0.3.7 pyhd8ed1ab_0 conda-forge