@inproceedings{ashkboos2024quarot,
author = {Ashkboos, Saleh and Mohtashami, Amirkeivan and Croci, Maximilian L. and Li, Bo and Jaggi, Martin and Alistarh, Dan and Hoefler, Torsten and Hensman, James and Cameron, Pashmina},
title = {QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMs},
booktitle = {2024 Neural Information Processing Systems},
year = {2024},
month = {March},
abstract = {We introduce QuaRot, a new Quantization scheme based on Rotations, which is able to quantize LLMs end-to-end, including all weights, activations, and KV cache in 4 bits. QuaRot rotates LLMs in a way that removes outliers from the hidden state without changing the output, making quantization easier. This computational invariance is applied to the hidden state (residual) of the LLM, as well as to the activations of the feed-forward components, aspects of the attention mechanism and to the KV cache. The result is a quantized model where all matrix multiplications are performed in 4-bits, without any channels identified for retention in higher precision. Our quantized LLaMa2-70B model has losses of at most 0.29 WikiText-2 perplexity and retains 99% of the zero-shot performance. Code is available at: https://github.com/spcl/QuaRot.},
url = {http://approjects.co.za/?big=en-us/research/publication/quarot-outlier-free-4-bit-inference-in-rotated-llms/},
note = {Preprint},
}