HybridCrypt-LLM: Lightweight privacy for LLM training and inference

Key Points

• Enhanced privacy mechanisms enable secure large language model training, ensuring data confidentiality.
• Hybrid cryptography solutions provide a lightweight approach to maintain security without heavy computational costs.
• Evaluation conducted includes various protocols aimed at increasing privacy during inference phases of LLM usage.
• Findings indicate that lightweight solutions are vital for practical applications, supporting broader adoption in sensitive contexts.