langgraph/how-tos/docs/quickstart/

[giscus[bot]]

langgraph/how-tos/docs/quickstart/

Build language agents as graphs

https://langchain-ai.github.io/langgraph/how-tos/docs/quickstart/

[Orbifold]

Despite best efforts, this code always gives a 'not JSON serializable' error. Often related to ModelMetaClass. Changing pydantic to langchain_core.pydantic_v1 does not help. Somewhere the custom tool is expected to get serialized and fails.

[hwchase17]

I just ran this in a fresh google collab without any issues. What is your environment/what is the full stack trace?

[Luo-hongyi]

I found this error too when I tried to make a custom tool which uses gpt-4o to summarize an image. It happened when the messages param receive an incorrect format json (I missed a tab somewhere). After formatting the code, the problem is solved. I guessed this error happened when the program try to turn something you write (which is not in a json format) into a json as param.

[KVSSetty]

For me it ran very smoothly with no errors and I am using Jupyter notebook on my local system.
But for my question:

runnable.invoke(HumanMessage("How can you help me?"))
It answered somewhat tricky using the same multiply tool like this:

[HumanMessage(content='what is the squareroot of 625', id='7b1816c3-ab9c-44a1-9524-5f2fcc480111'),
AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_oKDikke3w1I10TS0jJuX3bCs', 'function': {'arguments': '{"first_number":25,"second_number":25}', 'name': 'multiply'}, 'type': 'function'}]}, response_metadata={'token_usage': {'completion_tokens': 19, 'prompt_tokens': 58, 'total_tokens': 77}, 'model_name': 'gpt-3.5-turbo', 'system_fingerprint': None, 'finish_reason': 'tool_calls', 'logprobs': None}, id='run-3808af44-f86c-4f9c-aab9-790226452a30-0', tool_calls=[{'name': 'multiply', 'args': {'first_number': 25, 'second_number': 25}, 'id': 'call_oKDikke3w1I10TS0jJuX3bCs'}]),
ToolMessage(content='625', name='multiply', id='4d2ae1e9-6f7a-4408-bcf4-b5e0b196bd43', tool_call_id='call_oKDikke3w1I10TS0jJuX3bCs')]

My question is should I write a tool ,say "squareroot" tool, to get a straight forward answer?

[KVSSetty]

Sorry my invoke message is not what mentioned above , it is:

runnable.invoke(HumanMessage("what is the squareroot of 625"))