C:\Users\david\dev\trovono.venv\Lib\site-packages\langchain_core_api\deprecation.py:25: UserWarning: Core Pydantic V1 functionality isn't compatible with Python 3.14 or greater. from pydantic.v1.fields import FieldInfo as FieldInfoV1 Traceback (most recent call last): File "C:\Users\david\dev\trovono.venv\Lib\site-packages\langchain\agents\structured_output.py", line 410, in parse data = json.loads(raw_text) File "C:\Users\david\AppData\Roaming\uv\python\cpython-3.14.0-windows-x86_64-none\Lib\json_init_.py", line 346, in loads return _default_decoder.decode(s) ~~~~~~~~~~~~~~~~~~~~~~~^^^ File "C:\Users\david\AppData\Roaming\uv\python\cpython-3.14.0-windows-x86_64-none\Lib\json\decoder.py", line 348, in decode raise JSONDecodeError("Extra data", s, end) json.decoder.JSONDecodeError: Extra data: line 1 column 28 (char 27)

The above exception was the direct cause of the following exception:

Traceback (most recent call last): File "C:\Users\david\dev\trovono\repro.py", line 23, in <module> ).parse(message) ~~~~~^^^^^^^^^ File "C:\Users\david\dev\trovono.venv\Lib\site-packages\langchain\agents\structured_output.py", line 417, in parse raise ValueError(msg) from e ValueError: Native structured output expected valid JSON for MyModel, but parsing failed: Extra data: line 1 column 28 (char 27).

Fix Plan

To fix the issue, we need to modify the parse method in ProviderStrategyBinding to handle the case where the OpenAI model responds with multiple items of type 'text'. We can achieve this by iterating over the content of the AIMessage and parsing each item separately.

Here are the steps to fix the issue:

• Modify the parse method to iterate over the content of the AIMessage.
• For each item, check if it's a dictionary with a 'type' key equal to 'text' and a 'text' key that contains a valid JSON string.
• If the item is a valid JSON string, parse it into a Python object using the json.loads method.
• If the parsed object conforms to the expected schema (in this case, MyModel), return it as the result of the parse method.

Here's an example of how the modified parse method could look:

import json
from langchain.agents.structured_output import ProviderStrategyBinding
from langchain_core.messages import AIMessage
from pydantic import BaseModel

class MyModel(BaseModel):
    some_field: str

class CustomProviderStrategyBinding(ProviderStrategyBinding):
    def parse(self, message: AIMessage):
        for item in message.content:
            if item.get('type') == 'text' and item.get('text'):
                try:
                    data = json.loads(item['text'])
                    if isinstance(data, dict) and 'some_field' in data:
                        return MyModel(**data)
                except json.JSONDecodeError:
                    pass
        raise ValueError("Native structured output expected valid JSON for MyModel, but parsing failed")

# Usage
message = AIMessage(
    content=[
        dict(type="text", phase="commentary", text='{"some_field": "some text"}'),
        dict(type="text", phase="final_answer", text='{"some_field": "some other text"}'),
    ]
)

binding = CustomProviderStrategyBinding(
    schema=MyModel,
    schema_kind="pydantic",
)
result = binding.parse(message)
print(result)

Verification

To verify that the fix worked, you can run the modified code with the provided example AIMessage and check that the parse method returns the expected result. You can also test the code with different inputs to ensure that it handles various scenarios correctly.

Extra Tips

• Make sure to handle any potential exceptions that may occur during the parsing process, such as JSONDecodeError.
• Consider adding additional logging or debugging statements to help diagnose any issues that may arise during the parsing process.
• If you're using a custom schema, ensure that it's correctly defined and that the parse method is able to handle any nuances of the schema.