Inside Legion

Legion (formerly Yurts) delivers secure, orchestrated AI built for mission-critical workflows. Explore how we're empowering teams with intelligent, human-centered solutions for demanding environments.

Explore the build vs. buy dilemma in generative AI. Learn why adopting a hybrid strategy balances unique value and rapid deployment. Discover how businesses can achieve success by integrating Legion into their operations for both flexibility and immediate capabilities.

Discover how to secure generative AI without sacrificing flexibility. Learn about Legion's modular approach, which combines privacy-first design, adaptable infrastructure, and enterprise-grade security.

Legion integrates NVIDIA NIM microservices into its GenAI platform, expanding enterprise-ready LLMs, ensuring secure and flexible deployments. This enhances GPU-accelerated model variety, aiding secure, scalable AI solutions for enterprises.

Discover how Legion's RAG system matches Anthropic’s Contextual Retrieval performance on the Codebases dataset at just 1/300th of the cost. Explore our evaluation to see why Legion RAG is the smart, cost-effective choice for high-performance AI solutions.

Comparing RAG systems and LCW models on Needle in a Haystack benchmarks, showing RAG's superior performance and scalability, highlighting the need for better benchmarks for LCW models.

Next-gen AI without breaking the bank! AWQ, a quantization method, boosts deploying LLMs' cost-effectiveness by cutting GPU needs, enabling wider access to advanced AI technology at lower costs.

Enterprises need more than a powerful AI model; they require a secure, efficient, and compliant platform, not just disparate tools, to deploy generative AI at scale.

Discover why Legion is essential for businesses adopting generative AI, offering unmatched security, smooth integrations, and powerful analytics.

Learn to fine-tune LLMs with FIP & LoRA methods to beat "Catastrophic Forgetting" for robust AI applications across industries.

TL;DR: We benchmarked various open-source LLMs, including Llama-v2-7b-chat, finding they hallucinate around 55% of the time in context-aware Q&A tasks without tuning.