Summary of Moe-pruner: Pruning Mixture-of-experts Large Language Model Using the Hints From Its Router, by Yanyue Xie et al.
MoE-Pruner: Pruning Mixture-of-Experts Large Language Model using the Hints from Its Router
by Yanyue Xie, Zhi Zhang, Ding Zhou, Cong Xie, Ziang Song, Xin Liu, Yanzhi Wang, Xue Lin, An Xu
First submitted to arxiv on: 15 Oct 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 MoE-Pruner method prunes weights in Mixture-of-Experts (MoE) architectures by multiplying smallest magnitude weights by input activations and router weights. This one-shot pruning method does not require retraining or weight updates, making it efficient. Evaluations on Mixtral-8x7B and Mixtral-8x22B show that MoE-Pruner outperforms state-of-the-art Large Language Model (LLM) pruning methods. Additionally, the pruned models can benefit from expert-wise knowledge distillation with a pre-trained teacher model, maintaining performance after pruning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a way to make Mixture-of-Experts models more efficient by removing some of the extra information they contain. This is important because these models are very good at doing certain tasks, but they use a lot of memory and computing power. The new method, called MoE-Pruner, does this without needing to retrain or update the model. It also helps the pruned model learn from another model that has already been trained. The results show that using this method doesn’t harm the model’s performance and can even make it better. |
Keywords
» Artificial intelligence » Knowledge distillation » Large language model » Mixture of experts » One shot » Pruning » Teacher model
