Meta: LlamaGuard 2 8B

Meta: LlamaGuard 2 8B, a safeguard model based on the Llama 3 family, demonstrates exceptional speed and cost efficiency. It consistently ranks among the fastest models and offers highly competitive pricing across various benchmarks. Designed for prompt and response classification, it indicates safety status and identifies violated content categories. While excelling in its core function as a safeguard model, its performance on general-purpose benchmarks like Coding, Instruction Following, Reasoning, and Ethics is notably low, with 0.0% accuracy across these categories. The Email Classification benchmark showed a minimal 2.0% accuracy. This indicates that LlamaGuard 2 8B is highly specialized for its safety classification task and is not intended for general generative AI applications. Its strength lies in its dedicated safety function, where it has shown strong performance in human evaluations against leading closed-source models.