Summary of Qjl: 1-bit Quantized Jl Transform For Kv Cache Quantization with Zero Overhead, by Amir Zandieh et al.
QJL: 1-Bit Quantized JL Transform for KV Cache Quantization with Zero Overhead
by Amir Zandieh, Majid Daliri, Insu Han
First submitted to arxiv on: 5 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Performance (cs.PF)
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 presents a novel quantization approach called QJL, which addresses the memory requirements of Key-Value (KV) embeddings in large language models (LLMs). By leveraging Johnson-Lindenstrauss (JL) transform followed by sign-bit quantization, QJL eliminates the need for storing quantization constants, resulting in significant memory savings. The authors propose an asymmetric estimator for inner product computation and demonstrate its effectiveness on various LLMs and NLP tasks. When applied to quantize KV cache to 3 bits, QJL achieves a fivefold reduction in memory usage without compromising accuracy, while also speeding up runtime. This work has the potential to enable larger language models that can be deployed on limited-memory devices. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about finding ways to make large language models use less memory. Language models need to store lots of information, which takes up a lot of space. The authors came up with a new way to compress this information called QJL. It works by using a special transformation and then reducing the size of the numbers. This helps to save memory without losing accuracy. They tested it on different language models and tasks and found that it worked well, even when only 3 bits were used to store the information. This is important because it means we can use larger language models on devices with limited memory. |
Keywords
» Artificial intelligence » Nlp » Quantization
