Python: Bug: Getting http 400 error in assistant_agent_file_manipulation.py

[mrochon]

Describe the bug
Running the code on my own machine, with no code change except change to path for the csv input file, I am getting:
Error code: 400 - {'error': {'message': "Unknown parameter: 'tool_resources'.", 'type': 'invalid_request_error', 'param': 'tool_resources', 'code': 'unknown_parameter'}} on AzureAssistantAgent create.

To Reproduce
Steps to reproduce the behavior:
Running code as is.

Expected behavior
Some valid output, not an error.

Screenshots
If applicable, add screenshots to help explain your problem.

Platform

• Language: Python
• Source: python/samples/concepts/agents/assistant_agent/assistant_agent_file_manipulation.py
• AI model: AzureOpenAI:GPT-4o-mini(2024-07-18)
• IDE: VS Code
• OS: Windows

Additional context
Add any other context about the problem here.

[moonbox3]

Hello @mrochon, what Azure api version are you using? I am using AZURE_OPENAI_API_VERSION="2024-12-01-preview" and I don't have any 400s, it is running well for me. Looks like your AI model is fine, but please verify the api version. Also, what version of Python are you using? 400s are client-side errors, so there is some issue with your current configuration.

[mrochon]

My .env had

AZURE_OPENAI_API_VERSION="2024-02-15-preview"

After I changed it 2024-12-01-preview the code worked fine.

Nowhere in the source is that environment variable loaded though. Which class loads it behind the scenes?

[moonbox3]

@mrochon we use Pydantic settings, which loads env vars if present. When creating the assistant, we get a hold of the necessary settings. You can always override via the constructor, if necessary.

I will pick up a work item soon to improve the docs about which api versions and models are needed to run the assistants. Azure OpenAI is tricky because of all of the api versions.