Llama Guard 3 is a Llama-3.1-8B pretrained model, fine-tuned for content safety classification. Similar to previous versions, it can be used to classify content in both LLM inputs (prompt classification) and in LLM responses (response classification). It acts as an LLM – it generates text in its output that indicates whether a given prompt or response is safe or unsafe, and if unsafe, it also lists the content categories violated.
Fine-tuningDocs | Llama Guard 3 8B can be customized with your data to improve responses. Fireworks uses LoRA to efficiently train and deploy your personalized model |
On-demand DeploymentDocs | On-demand deployments give you dedicated GPUs for Llama Guard 3 8B using Fireworks' reliable, high-performance system with no rate limits. |
Llama Guard 3 8B is a content safety classification model developed by Meta, based on the Llama 3.1 8B architecture. It is fine-tuned to classify prompts and responses as safe or unsafe across 14 hazard categories based on the MLCommons taxonomy. It supports multilingual content moderation and is designed to integrate with systems using Llama 3.1.
The model supports a context length of 131,072 tokens (131.1K).
Fireworks supports the full 131.1K token context window on on-demand deployments.
Yes. Meta provides a quantized INT8 variant that reduces checkpoint size by ~40% with minimal performance degradation. Performance remains comparable across key benchmarks.
The maximum output is constrained by the 131.1K token context limit (prompt + output). The model typically generates short classification strings.
The model contains 8 billion parameters.
Yes. Fireworks supports LoRA-based fine-tuning using its Reserved Fine-Tuning (RFT) infrastructure.
Token usage is measured as prompt + output, and constrained by the 131.1K token context window.
Llama Guard 3 8B is released under the Llama 3 Community License (Meta). This license restricts certain commercial use cases and requires user agreement for access. Full terms are available at https://llama.meta.com/license.
