Fine-Tuned LLMs for Defense
LLMs customized specifically for the unique missions of the U.S. national security community, now available in Scale Donovan.
Purpose-built models for defense and intelligence communities.
Available exclusively in controlled U.S. government environments on Scale Donovan, custom LLMs for defense are built on Meta’s LLama 3. They enable national security users to apply generative AI to defense-related problem sets:
Military and Intelligence Operations Planning
Defense LLM empowers service members and national security professionals to use generative AI to understand adversary vulnerabilities and effectively plan military or intelligence operations.
Target Analysis
Defense LLM allows military planners to understand how an adversary could plan an attack against a U.S. military base and explore possible ways to counter such an attack.
Fine-tuned and evaluated for the tone and precise language that national security professionals require.
Using Scale’s proprietary Data Engine, custom LLMs for defense applications are tuned then tested and evaluated to ensure model responses meet performance, reliability, and safety requirements.
Supervised Fine-Tuning & RLHF
Using data generated from our experts, LLMs programmed for Defense have DoD domain-specific knowledge and use a writing style that adheres to the Office of the Director of National Intelligence (ODNI) guidelines. Customized models for defense are able to provide accurate, meaningful, and relevant responses.
Test & Evaluation
Custom LLMs for Defense are tested with the first Scale-created Department of Defense-specific evaluation benchmarks to ensure that the model understands domain knowledge and government terminology.