How can we access the prompt given to the LLM?

[Chasearmer]

As the title suggests, how can we access the prompt that is generated from the Pydantic response model and passed to the LLM? I would be very interested in seeing what that looks, especially towards accessing the clarity of prompts created from nested Pydantic models.

[jxnl]

Making some improvements to docs

https://jxnl.github.io/instructor/concepts/logging/

[Mr-Ruben]

if you set the logging level to debug openai should print everything!

This is rather unclear/unspecific.

• Set it where? ( call / module )
• Of who? (instructor / openai )
• How? (not everyone is familiar with python logging module)

---

This got me on the right track
https://stackoverflow.com/questions/76256249/logging-in-the-open-ai-python-library

And after some experimentation (I have a nested logging setup)

Get my top logger (this may be different for you)

toplog = logging.getLogger("<MyModuleName>")
toplog.handlers  # This should show something

This hooks openai logs to my own ones

openai_loggers = [
    logger for logger in loggers if logger.name.startswith("openai")]

for openai_logger in openai_loggers:
    # Get the openai logger
    openai_logger = logging.getLogger(openai_logger.name)

    # Set the openai logger level to DEBUG
    openai_logger.setLevel(logging.DEBUG)

    # Add your logger's handlers to the openai logger
    for handler in toplog.handlers:
        openai_logger.addHandler(handler)

On the logs you will see the full body of the call

Extracting and formatting

'method': 'post', 
'url': '/chat/completions', 
'files': None, 
'json_data': {'messages': [{'role': 'user', 'content': 'Extract Jason is 25 years old'}], 
'model': 'gpt-3.5-turbo', 
'function_call': {'name': 'UserDetail'}, 
'functions': [
{'name': 'UserDetail', 'description': 'Correctly extracted `UserDetail` with all the required parameters with correct types', 
'parameters': {'properties': {'name': {'title': 'Name', 'type': 'string'}, 				
                                            'age': {'title': 'Age', 'type': 'integer'}}, 
'required': ['age', 'name'], 
'type': 'object'}}], 
'max_tokens': 1500, 'n': 1, 'temperature': 0.1}

https://i.imgur.com/GcYedZV.png

I hope this saves time to the next one.

[leobeeson]

Thank you, @Mr-Ruben! I managed to get a working answer on the Instructor Discord here, but your suggestion is even more straightforward (for those of us not familiar with python logging module). Thanks, again.

[jxnl]

Is anyone willing to help add this to docs? I'll make an issue

[leobeeson]

I'll take it! It'll be a great Python logging learning opportunity. Please assign the issue to me. Thanks.

[jxnl]

ty ty!