Summary of Distillm: Towards Streamlined Distillation For Large Language Models, by Jongwoo Ko et al.
DistiLLM: Towards Streamlined Distillation for Large Language Models
by Jongwoo Ko, Sungnyun Kim, Tianyi Chen, Se-Young Yun
First submitted to arxiv on: 6 Feb 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
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 proposed DistiLLM framework is a more effective and efficient knowledge distillation (KD) method for compressing large language models, reducing inference costs and memory footprint while preserving capabilities. It comprises two components: a novel skew Kullback-Leibler divergence loss and an adaptive off-policy approach to leverage student-generated outputs. The method outperforms recent KD methods in building high-performing student models, achieving up to 4.3 times speedup. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A team of researchers created a new way to make large language models smaller and faster. They call it DistiLLM. It helps by teaching the model to be more efficient with its output, which makes it run quicker without losing performance. This is important because these big models are very useful but can take up lots of computer power and storage space. The new method was tested on different tasks and showed great results. |
Keywords
* Artificial intelligence * Inference * Knowledge distillation
