ai-agents-vllm-gptoss-null-con-using-gpt-oss-harmony-models-with-vllm-0-13-0-in-m-db0122c0

THEN Downgrade vLLM to version 0.10.1 or 0.11.2. Alternatively, if you cannot downgrade, rewrite prior assistant messages in the chat history as user messages with a prefix like '[ASSISTANT message]:' to avoid the bug. Note that this workaround also applies to large user prompts (30k+ tokens) that may trigger the same null-content issue.

# Workaround: change role of prior assistant messages
chat = [
    {"role": "system", "content": "You are a helpful assistant!"},
    {"role": "user", "content": "What is 2 + 5?"},
    {"role": "user", "content": "[ASSISTANT message]:\n\n6, of course!"},
    {"role": "user", "content": "No, try again. Respond with {'response': '<your answer>'}"}
]