Why does my Python API client send extra requests after a single failure?

I work on CloudGPU. Check for two retry owners: the SDK and your application. A single call in your code can become several HTTP attempts if both layers retry. In the OpenAI Python SDK, max_retries=0