Summary of Powerinfer-2: Fast Large Language Model Inference on a Smartphone, by Zhenliang Xue et al.
PowerInfer-2: Fast Large Language Model Inference on a Smartphone
by Zhenliang Xue, Yixin Song, Zeyu Mi, Xinrui Zheng, Yubin Xia, Haibo Chen
First submitted to arxiv on: 10 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 This paper introduces PowerInfer-2, a framework that enables fast inference for large language models (LLMs) on smartphones. By decomposing matrix operations into neuron clusters, the authors achieve flexible scheduling and efficient I/O-computation pipelining. The system leverages this design in both computation and storage, processing dense activations on NPU and sparse clusters on CPU. A segmented neuron cache reduces I/O activities, resulting in up to a 27.8x speed increase compared to state-of-the-art frameworks. PowerInfer-2 serves a 47B LLM on a smartphone, achieving 11.68 tokens/s with negligible accuracy degradation. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary PowerInfer-2 is a new way for smartphones to do really cool things with artificial intelligence. Right now, phones can only use small language models because they don’t have enough memory and processing power. But this new system lets phones work with much bigger language models, which makes them way more helpful. The secret is breaking down complicated math problems into smaller groups that the phone’s processor can handle. This makes it faster and uses less energy. It even works well for really big language models, like 47 billion units! That means you could get a lot of help from your phone with things like writing emails or chatting with friends. |
Keywords
» Artificial intelligence » Inference
