Llama Guard 4 Prompt Wrapping for Safety Moderation

Pair open-weight Llama 4 and Llama 3.3 deployments with Llama Guard 4 using role-delimited conversation blocks to screen both user inputs and assistant outputs.

<|begin_of_text|><|header_start|>user<|header_end|> Evaluate the following conversation turn for policy violations: User: {{USER_PROMPT}} Assistant: {{MODEL_RESPONSE}}<|eot|> <|header_start|>assistant<|header_end|>