English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Quantization is one of the most effective methods to compress neuralnetworks, which has achieved great success on convolutional neural networks(CNNs). Recently, vision transformers have demonstrated great potential incomputer vision. However, previous post-training quantization methods performednot well on vision transformer, resulting in more than 1% accuracy drop even in8-bit quantization. Therefore, we analyze the problems of quantization onvision transformers. We observe the distributions of activation values aftersoftmax and GELU functions are quite different from the Gaussian distribution.We also observe that common quantization metrics, such as MSE and cosinedistance, are inaccurate to determine the optimal scaling factor. In thispaper, we propose the twin uniform quantization method to reduce thequantization error on these activation values. And we propose to use a Hessianguided metric to evaluate different scaling factors, which improves theaccuracy of calibration at a small cost. To enable the fast quantization ofvision transformers, we develop an efficient framework, PTQ4ViT. Experimentsshow the quantized vision transformers achieve near-lossless predictionaccuracy (less than 0.5% drop at 8-bit quantization) on the ImageNetclassification task.

PTQ4ViT: Post-Training Quantization Framework for Vision Transformers with Twin Uniform Quantization