English

Integer Scale: A Free Lunch for Faster Fine-grained Quantization of LLMs

Machine Learning 2024-05-29 v2 Artificial Intelligence

Abstract

We introduce Integer Scale, a novel post-training quantization scheme for large language models that effectively resolves the inference bottleneck in current fine-grained quantization approaches while maintaining similar accuracies. Integer Scale is a free lunch as it requires no extra calibration or fine-tuning which will otherwise incur additional costs. It can be used plug-and-play for most fine-grained quantization methods. Its integration results in at most 1.85x end-to-end speed boost over the original counterpart with comparable accuracy. Additionally, due to the orchestration of the proposed Integer Scale and fine-grained quantization, we resolved the quantization difficulty for Mixtral-8x7B and LLaMA-3 models with negligible performance degradation, and it comes with an end-to-end speed boost of 2.13x, and 2.31x compared with their FP16 versions respectively.

Cite

@article{arxiv.2405.14597,
  title  = {Integer Scale: A Free Lunch for Faster Fine-grained Quantization of LLMs},
  author = {Qingyuan Li and Ran Meng and Yiduo Li and Bo Zhang and Yifan Lu and Yerui Sun and Lin Ma and Yuchen Xie},
  journal= {arXiv preprint arXiv:2405.14597},
  year   = {2024}
}
R2 v1 2026-06-28T16:37:19.742Z