From Pre-training to Deployment: A Review of Large Language Model Technologies and Applications

Abstract

Large language models have reshaped natural language processing and are increasingly used as general interfaces for knowledge work. This review summarizes their architectural foundations, pre-training and post-training methods, application patterns, evaluation practices, and governance challenges. The literature shows that progress depends not only on parameter scale, but also on data quality, compute allocation, instruction tuning, preference optimization, retrieval augmentation, and deployment discipline. The paper also discusses hallucination, bias, privacy, cost, and safety risks, and identifies future directions in multimodal modeling, agentic systems, domain adaptation, and trustworthy governance.