Summary of Factorllm: Factorizing Knowledge Via Mixture Of Experts For Large Language Models, by Zhongyu Zhao et al.
FactorLLM: Factorizing Knowledge via Mixture of Experts for Large Language Models
by Zhongyu Zhao, Menghang Dong, Rongyu Zhang, Wenzhao Zheng, Yunpeng Zhang, Huanrui Yang, Dalong Du, Kurt Keutzer, Shanghang Zhang
First submitted to arxiv on: 15 Aug 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 A novel approach to Large Language Models (LLMs) is proposed, which decomposes well-trained dense Feed-Forward Networks (FFNs) into sparse sub-networks without requiring any further modifications. This factorization enables efficient knowledge activation and accelerates computational processes. The FactorLLM model combines the Mixture-of-Experts (MoE) router with a Prior-Approximate (PA) loss term to facilitate dynamic expert activation and knowledge adaptation. Experimental results demonstrate the effectiveness of FactorLLM, achieving comparable performance to the source model while increasing inference speed by up to 30%. This approach is particularly useful for LLMs, as it minimizes computational overhead. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large Language Models (LLMs) are super smart computers that can learn lots of things. Right now, these models use a special type of network called Feed-Forward Networks (FFNs). These networks help the model learn and remember lots of different facts and language rules. But sometimes, these networks can get too big and confusing, which slows them down. To solve this problem, researchers have developed a new way to break down these networks into smaller, more efficient parts. This is called FactorLLM, and it’s really good at quickly finding the right information and using it to answer questions. |
Keywords
» Artificial intelligence » Inference » Mixture of experts
