Integration broken for non-Tier 3 OpenAI accounts due to o3-mini requirement

[SublimePotato]

Git provider (optional)

Azure

System Info (optional)

codiumai/pr-agent in Azure DevOps with the default .yml

Issues details

The recent commit broke the integration for me since the OpenAI account I am using is not yet Tier 3. This is an easy fix, but due to OpenAI's tier handling, I have to wait 7 days.

Error message:
pr_agent.algo.ai_handlers.litellm_ai_handler:chat_completion:276 - Error during LLM inference: litellm.NotFoundError: OpenAIException - Error code: 404 - {'error': {'message': 'The model o3-mini does not exist or you do not have access to it.', 'type': 'invalid_request_error', 'param': None, 'code': 'model_not_found'}}

Additionally, the fallback model doesn't seem to be used and another change seems to have made the previous default incompatible. Locking manually to gpt-4o-2024-11-20 results in the following error:

pr_agent.algo.ai_handlers.litellm_ai_handler:chat_completion:276 - Error during LLM inference: litellm.APIError: AzureException APIError - argument of type 'NoneType' is not iterable

This issue will resolve itself once o3-mini becomes available for this account next week. However, it might confuse new users that OpenAI Tier 3 API access is mandatory in the default configuration.

[ofir-frd]

You can modify the configuration to use any model you have access to while waiting for your Tier 3 access to activate. For immediate compatibility, you can set the model to Claude 3.7 Sonnet or GPT-4o (if you have access) in your configuration file. This would bypass the requirement for the o3-mini model that's currently causing the error.

Regarding the fallback model issue, the NoneType is not iterable error suggests there might be a parameter initialization problem when using certain model strings with Azure. You might want to try a different fallback model format, for example. We would appreciate your contribution for investigating this further.