Summary of Slim: One-shot Quantization and Sparsity with Low-rank Approximation For Llm Weight Compression, by Mohammad Mozaffari et al.
SLiM: One-shot Quantization and Sparsity with Low-rank Approximation for LLM Weight Compression
by Mohammad Mozaffari, Amir Yazdanbakhsh, Maryam Mehri Dehnavi
First submitted to arxiv on: 12 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); 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 introduces SLIM, a novel one-shot compression framework for large language models (LLMs). Unlike conventional methods that require computationally expensive retraining, SLIM eliminates this cost by holistically integrating hardware-friendly quantization, sparsity, and low-rank approximation. The approach starts with probabilistic quantization (SLIM-Quant) followed by semi-structured sparsity pruning. To compensate for introduced errors, a novel saliency function is used to compute the value of low-rank adapters. SLIM achieves up to 5.66% accuracy improvement for LLaMA-2-7B with 4-bit weight quantization and 2:4 sparsity, outperforming prior methods. Compressed models show layer-wise speedup on Nvidia GPUs, with optional PEFT recipe improving accuracy by up to 1.66%. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper makes it possible to make large language models smaller without losing their ability to understand text. The current way of compressing these models is slow and expensive, but this new method called SLIM can do it quickly and accurately. It works by combining three techniques: making the numbers in the model smaller (quantization), removing some parts of the model (sparsity), and adjusting the remaining parts to make up for the loss. This approach improves the accuracy of the compressed models by up to 5.66% compared to previous methods. |
Keywords
» Artificial intelligence » Llama » One shot » Pruning » Quantization
