CODE GENERATION WITH LARGE LANGUAGE MODELS: AUTO-COMPLETION AND BUG FIXING

This paper presented a framework for code generation using large language models for auto-completion and automated bug fixing. Transformer-based architectures demonstrate strong capabilities in understanding programming patterns and generating syntactically valid and context-aware code.   Future work includes integrating reinforcement learning from human feedback to improve reliability, combining LLM outputs with static code analyzers for security validation, optimizing models for edge deployment, and enhancing explainability of generated solutions.

[1]             A. Vaswani et al., “Attention Is All You Need,” Advances in Neural Information Processing Systems, 2017.   [2]             J. Devlin et al., “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,” NAACL, 2019.   [3]             M. Chen et al., “Evaluating Large Language Models Trained on Code,” arXiv, 2021.   [4]             Z. Feng et al., “CodeBERT: A Pre-Trained Model for Programming and Natural Languages,” EMNLP, 2020.   [5]             Y. Liu et al., “RoBERTa: A Robustly Optimized BERT Pretraining Approach,” arXiv, 2019.   [6]             T. Brown et al., “Language Models are FewShot Learners,” NeurIPS, 2020.   [7]             R. Tufano et al., “An Empirical Study on Learning Bug-Fixing Patches in the Wild,” IEEE Transactions on Software Engineering, 2019.   [8]             M. Allamanis et al., “Learning to Represent Programs with Graphs,” ICLR, 2018.   [9]             D. Bahdanau et al., “Neural Machine Translation by Jointly Learning to Align and Translate,” ICLR, 2015.   [10]          T. Brown et al., “Language Models are Few-Shot Learners,” NeurIPS, 2020.