ai-agents-structured-output-mu-using-gpt-oss-harmony-model-with-structured-output-17519e90

THEN Downgrade to vLLM 0.10.1 or earlier. Alternatively, in the message history, change the role of the assistant message to 'user' and prefix with '[ASSISTANT message]:\n\n' to bypass the bug. Avoid using structured outputs in multi-turn conversations with GPT-OSS models.

chat_extended = [
    {"role": "user", "content": "[ASSISTANT message]:\n\n6, of course!"},
    {'role': 'user', 'content': 'No, try again. Respond with {"response": "<your answer>"}'}
]