Use inference APIs for executing Llama Guard

[ashwinb]

We should use Inference APIs to execute Llama Guard instead of directly needing to use HuggingFace APIs. The actual inference consideration is handled by Inference.