infrastructure-azure-ad-token-provi-using-azure-ad-token-provider-in-litellm-s-complet-6c6b82d9

THEN Upgrade litellm to a version that includes PR #8468 (e.g., after 1.6.0, check release notes for fix). If upgrade is not possible, instantiate a new client per request or fall back to a static token. Alternatively, apply the patch from PR #8468 locally: ensure the azure_ad_token_provider callable is correctly forwarded to the underlying OpenAI client's AzureADTokenProvider configuration.

from litellm import Router
model_list = [
    {
        "model_name": "fast_model",
        "litellm_params": {
            "model": "azure/gpt-4o-mini",
            "base_url": "https://XXX.openai.azure.com/",
            "api_version": "2024-08-01-preview",
            "azure_ad_token_provider": generate_azure_token
        }
    }
]
router = Router(model_list=model_list)
response = router.completion("fast_model", messages=[{"role": "user", "content": "Hello"}])
# Before fix: error. After fix: works.