Llama 3.1 NemoGuard 8B TopicControl NIM#

About the Microservice#

The microservice serves a GPU-accelerated LLM model for conversational dialog moderation to keep conversations on-topic and build trustworthy LLM applications.

The model is a dialog moderation model trained by NVIDIA from a Llama 3.1 8B Instruct base model. The model is trained using the CantTalkAboutThis dataset – a carefully curated collection of 10 thousand dialogue samples – to detect and ensure user interactions are in line with developer-defined guidelines. The dataset has been released with a commercially friendly license.

In contrast with content moderation models such as Shield Gemma and Llama Guard 3, that rely on a pre-defined taxonomy of harms, this model enables developers to specify their own boundaries regarding allowed and disallowed subjects. This adaptability makes it a more versatile and general-purpose complement to content moderation models.