empirical study of six post-training quantization techniques (GPTQ, AWQ, QuIP#, AQLM, BitsAndBytes, GGUF) for code generation — measuring functional correctness, code quality, and robustness to prompt complexity
studying whether a single QLoRA-tuned model can jointly handle code generation, translation, and summarization while preserving functional correctness and non-functional code quality
leveraging parameter-efficient fine-tuning with QLoRA for resource-efficient code summarization
analyzing readability, maintainability, security and efficiency characteristics of code generated by quantized models
optimizing training datasets to improve code summarization model performance and efficiency
systematic literature review and implementation of PEFT techniques for large code models
a curated benchmark studying how the natural language of a prompt (Chinese, Hindi, Spanish, Italian vs. English) affects the correctness and quality of code generated by GPT-4o mini, DeepSeek, and Claude
