Parameter-Efficient Instruction Tuning Code Large Language Models: A Comprehensive Study on OctoCoder Models
Abstract: The high cost of full-parameter fine-tuning (FFT) of Large Language Models (LLMs) has led to a series of parameter-efficient fine-tuning (PEFT) methods. However, it remains unclear which methods provide the best cost-performance trade-off at different model scales. We introduce Astraios, a suite of 28 instruction-tuned OctoCoder models using 7 tuning methods and 4 model sizes up to 16 billion parameters. Through investigations across 5 tasks and 8 different datasets encompassing both code comprehension and code generation tasks, we find that FFT generally leads to the best downstream performance across all scales, and PEFT methods differ significantly in their efficacy based on the model scale. LoRA usually offers the most favorable trade-off between cost and performance. Further investigation into the effects of these methods on both model robustness and code security reveals that larger models tend to demonstrate reduced robustness and less security. At last, we explore the relationships among updated parameters, and task performance. We find that the tuning effectiveness observed in small models generalizes well to larger models, and the validation loss in instruction tuning can be a reliable indicator of overall downstream performance. We believe that our findings of PEFT can generalize to other decoder-only LLMs.
Paper Type: Long
Research Area: Special Theme (conference specific)
Research Area Keywords: parameter-efficient-training, code generation and understanding, empirical study
Contribution Types: Model analysis & interpretability, NLP engineering experiment, Publicly available software and/or pre-trained models
Languages Studied: English, Python, Java, C
Submission Number: 2450
Loading