SKIM: Any-bit Quantization Pushing The Limits of Post-Training Quantization

Runsheng Bai, Bo Liu, Qiang Liu
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:2520-2534, 2025.

Abstract

Large Language Models (LLMs) exhibit impressive performance across various tasks, but deploying them for inference poses challenges. Their high resource demands often necessitate complex, costly multi-GPU pipelines, or the use of smaller, less capable models. While quantization offers a promising solution utilizing lower precision for model storage, existing methods frequently experience significant performance drops at lower precision levels. Additionally, they typically provide only a limited set of solutions at specific bit levels, many of which are extensively manually tuned. To address these challenges, we propose a new method called SKIM: Scaled K-means clustering wIth Mixed precision. Our approach introduces two novel techniques: 1. A greedy algorithm to solve approximately optimal bit allocation across weight channels, and 2. A trainable scaling vector for non-differentiable K-means clustering. These techniques substantially improve the model performance and can be adapted to any given bit. Notably, in terms of perplexity, our method narrows the gap between quantized LLaMA models and their full precision counterparts by around 14% on average.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-bai25c, title = {{SKIM}: Any-bit Quantization Pushing The Limits of Post-Training Quantization}, author = {Bai, Runsheng and Liu, Bo and Liu, Qiang}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {2520--2534}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/bai25c/bai25c.pdf}, url = {https://proceedings.mlr.press/v267/bai25c.html}, abstract = {Large Language Models (LLMs) exhibit impressive performance across various tasks, but deploying them for inference poses challenges. Their high resource demands often necessitate complex, costly multi-GPU pipelines, or the use of smaller, less capable models. While quantization offers a promising solution utilizing lower precision for model storage, existing methods frequently experience significant performance drops at lower precision levels. Additionally, they typically provide only a limited set of solutions at specific bit levels, many of which are extensively manually tuned. To address these challenges, we propose a new method called SKIM: Scaled K-means clustering wIth Mixed precision. Our approach introduces two novel techniques: 1. A greedy algorithm to solve approximately optimal bit allocation across weight channels, and 2. A trainable scaling vector for non-differentiable K-means clustering. These techniques substantially improve the model performance and can be adapted to any given bit. Notably, in terms of perplexity, our method narrows the gap between quantized LLaMA models and their full precision counterparts by around 14% on average.} }
Endnote
%0 Conference Paper %T SKIM: Any-bit Quantization Pushing The Limits of Post-Training Quantization %A Runsheng Bai %A Bo Liu %A Qiang Liu %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-bai25c %I PMLR %P 2520--2534 %U https://proceedings.mlr.press/v267/bai25c.html %V 267 %X Large Language Models (LLMs) exhibit impressive performance across various tasks, but deploying them for inference poses challenges. Their high resource demands often necessitate complex, costly multi-GPU pipelines, or the use of smaller, less capable models. While quantization offers a promising solution utilizing lower precision for model storage, existing methods frequently experience significant performance drops at lower precision levels. Additionally, they typically provide only a limited set of solutions at specific bit levels, many of which are extensively manually tuned. To address these challenges, we propose a new method called SKIM: Scaled K-means clustering wIth Mixed precision. Our approach introduces two novel techniques: 1. A greedy algorithm to solve approximately optimal bit allocation across weight channels, and 2. A trainable scaling vector for non-differentiable K-means clustering. These techniques substantially improve the model performance and can be adapted to any given bit. Notably, in terms of perplexity, our method narrows the gap between quantized LLaMA models and their full precision counterparts by around 14% on average.
APA
Bai, R., Liu, B. & Liu, Q.. (2025). SKIM: Any-bit Quantization Pushing The Limits of Post-Training Quantization. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:2520-2534 Available from https://proceedings.mlr.press/v267/bai25c.html.

Related Material