SparseGPT: Massive Language Models Can be Accurately Pruned in One-Shot

Elias Frantar, Dan Alistarh
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:10323-10337, 2023.

Abstract

We show for the first time that large-scale generative pretrained transformer (GPT) family models can be pruned to at least 50% sparsity in one-shot, without any retraining, at minimal loss of accuracy. This is achieved via a new pruning method called SparseGPT, specifically designed to work efficiently and accurately on massive GPT-family models. We can execute SparseGPT on the largest available open-source models, OPT-175B and BLOOM-176B, in under 4.5 hours, and can reach 60% unstructured sparsity with negligible increase in perplexity: remarkably, more than 100 billion weights from these models can be ignored at inference time. SparseGPT generalizes to semi-structured (2:4 and 4:8) patterns, and is compatible with weight quantization approaches. The code is available at: https://github.com/IST-DASLab/sparsegpt.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-frantar23a, title = {{S}parse{GPT}: Massive Language Models Can be Accurately Pruned in One-Shot}, author = {Frantar, Elias and Alistarh, Dan}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {10323--10337}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/frantar23a/frantar23a.pdf}, url = {https://proceedings.mlr.press/v202/frantar23a.html}, abstract = {We show for the first time that large-scale generative pretrained transformer (GPT) family models can be pruned to at least 50% sparsity in one-shot, without any retraining, at minimal loss of accuracy. This is achieved via a new pruning method called SparseGPT, specifically designed to work efficiently and accurately on massive GPT-family models. We can execute SparseGPT on the largest available open-source models, OPT-175B and BLOOM-176B, in under 4.5 hours, and can reach 60% unstructured sparsity with negligible increase in perplexity: remarkably, more than 100 billion weights from these models can be ignored at inference time. SparseGPT generalizes to semi-structured (2:4 and 4:8) patterns, and is compatible with weight quantization approaches. The code is available at: https://github.com/IST-DASLab/sparsegpt.} }
Endnote
%0 Conference Paper %T SparseGPT: Massive Language Models Can be Accurately Pruned in One-Shot %A Elias Frantar %A Dan Alistarh %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-frantar23a %I PMLR %P 10323--10337 %U https://proceedings.mlr.press/v202/frantar23a.html %V 202 %X We show for the first time that large-scale generative pretrained transformer (GPT) family models can be pruned to at least 50% sparsity in one-shot, without any retraining, at minimal loss of accuracy. This is achieved via a new pruning method called SparseGPT, specifically designed to work efficiently and accurately on massive GPT-family models. We can execute SparseGPT on the largest available open-source models, OPT-175B and BLOOM-176B, in under 4.5 hours, and can reach 60% unstructured sparsity with negligible increase in perplexity: remarkably, more than 100 billion weights from these models can be ignored at inference time. SparseGPT generalizes to semi-structured (2:4 and 4:8) patterns, and is compatible with weight quantization approaches. The code is available at: https://github.com/IST-DASLab/sparsegpt.
APA
Frantar, E. & Alistarh, D.. (2023). SparseGPT: Massive Language Models Can be Accurately Pruned in One-Shot. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:10323-10337 Available from https://proceedings.mlr.press/v202/frantar23a.html.

Related Material