Summary of Astraios: Parameter-efficient Instruction Tuning Code Large Language Models, by Terry Yue Zhuo et al.
Astraios: Parameter-Efficient Instruction Tuning Code Large Language Models
by Terry Yue Zhuo, Armel Zebaze, Nitchakarn Suppattarachai, Leandro von Werra, Harm de Vries, Qian Liu, Niklas Muennighoff
First submitted to arxiv on: 1 Jan 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Software Engineering (cs.SE)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The Astraios suite is introduced as a set of 28 Large Language Models (LLMs) with varying sizes up to 16 billion parameters, utilizing 7 fine-tuning methods. Investigations across 5 tasks and 8 datasets demonstrate that full-parameter fine-tuning generally outperforms parameter-efficient approaches, with LoRA offering the best cost-performance trade-off. The study also explores model robustness and code security, revealing larger models tend to be less robust and secure. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large Language Models (LLMs) are being used in various tasks, but they can be very expensive to train. To solve this problem, many different methods have been tried to fine-tune the models. This study looks at 28 different models that use these methods and tests them on 5 different tasks with 8 different datasets. The results show that training all of a model’s parameters at once usually works best, but some other methods can be good too. The study also found that larger models are less robust and secure than smaller ones. |
Keywords
» Artificial intelligence » Fine tuning » Lora » Parameter efficient
