Astraios: Parameter-Efficient Instruction Tuning Code Large Language Models
This work provides empirical guidance for practitioners in AI and software engineering on selecting fine-tuning methods for code LLMs, though it is incremental as it builds on existing PEFT techniques.
The study tackled the problem of unclear cost-performance trade-offs in parameter-efficient fine-tuning (PEFT) methods for code large language models by evaluating 28 models across 7 methods and 4 sizes up to 16B parameters, finding that full-parameter fine-tuning generally performs best, LoRA offers the best trade-off, and larger models show reduced robustness and security.
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, cross-entropy loss, 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.