Summary of Low-rank Correction For Quantized Llms, by Meyer Scetbon et al.
Low-Rank Correction for Quantized LLMs
by Meyer Scetbon, James Hensman
First submitted to arxiv on: 10 Dec 2024
Categories
- Main: Machine Learning (stat.ML)
- Secondary: 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 We examine model compression for Large Language Models (LLMs) at post-training time, aiming to compress a well-trained model using minimal calibration input data. Our approach, dubbed low-rank correction, targets quantization errors in LLM activations by introducing full-precision low-rank weight matrices that operate on unquantized activation values. A joint optimization problem is formulated to jointly optimize the quantized weights and additional low-rank weight matrices for both weight and activation quantization (W4A4). Our method reduces the accuracy gap with the original model by more than 50% using ranks equivalent to 10% of the original weight matrix size, achieving complete accuracy closure at 30%. We evaluate our results on four recent LLMs: Llama-2, Llama-3, Phi-3, and Mixtral models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper talks about making Large Language Models (LLMs) smaller without losing their ability to understand language. They propose a new way to fix errors that happen when we simplify the model’s internal workings. By adding some extra information, they can make the model more accurate while still keeping it small. They tested this idea on four different LLMs and showed that it works really well. |
Keywords
* Artificial intelligence * Llama * Model compression * Optimization * Precision * Quantization
