Institutional Platform for Secure Self-Service Large Language Model Exploration.

Area of Science:

• Artificial Intelligence
• Biomedical Informatics
• Computer Science

Background:

• Large language models (LLMs) offer significant potential for scientific discovery and biomedical informatics.
• Current LLM accessibility is limited by complexity and cost.
• Customization of LLMs is crucial for specialized research applications.

Purpose of the Study:

• To introduce a user-friendly platform for creating customized large language models (LLMs).
• To enhance accessibility of advanced AI models for researchers.
• To provide secure and affordable LLM services.

Main Methods:

• Development of a platform leveraging multi-LoRA inference for efficient custom adapter accommodation.
• Implementation of a tenant-aware computational network using agent-based methods.
• Integration of dataset curation, model training, secure inference, and text-based feature extraction.
• Emphasis on process and data isolation, end-to-end encryption, and role-based resource authentication for security.

Main Results:

• A system architecture enabling efficient customization of LLMs.
• Secure and isolated computational resources for LLM services.
• Demonstration of a unified system from isolated resources.
• Facilitation of secure, affordable LLM services.

Conclusions:

• The developed platform simplifies access to cutting-edge AI, specifically large language models.
• It supports scientific discovery and the advancement of biomedical informatics through accessible AI.
• The system provides a secure and cost-effective solution for customized LLM deployment.