One-Shot Sensitivity-Aware Mixed Sparsity Pruning for Large Language Models
Authors:
- 14 April 2024
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 11296-11300
Abstract
Various Large Language Models (LLMs) from the Generative Pretrained Transformer (GPT) family have achieved outstanding performances in a wide range of text generation tasks. However, the enormous model sizes have hindered their practical use in real-world applications due to high inference latency. Therefore, improving the efficiencies of LLMs through quantization, pruning, and other means has been a key issue in LLM studies. In this work, we propose a method based on Hessian sensitivity-aware mixed sparsity pruning to prune LLMs to at least 50% sparsity without the need of any retraining. It allocates sparsity adaptively based on sensitivity, allowing us to reduce pruning-induced error while maintaining the overall sparsity level. The advantages of the proposed method exhibit even more when the sparsity is extremely high. Furthermore, our method is compatible with quantization, enabling further compression of LLMs.Keywords
All Related Versions
This publication has 16 references indexed in Scilit:
- TruthfulQA: Measuring How Models Mimic Human Falsehoods2022
- Block Pruning For Faster Transformers2021
- BinaryBERT: Pushing the Limit of BERT Quantization2021
- Meta-KD: A Meta Knowledge Distillation Framework for Language Model Compression across Domains2021
- Contrastive Distillation on Intermediate Representations for Language Model Compression2020
- Structured Pruning of Large Language Models2020
- HAWQ: Hessian AWare Quantization of Neural Networks With Mixed-Precision2019
- Randomized Algorithms for Matrices and Data2012
- Randomized algorithms for estimating the trace of an implicit symmetric positive semi-definite matrixJournal of the ACM, 2011
- PiQA: an algebra for querying protein data sets2003