CLIP

image-text similarity and for zero-shot image classification

LG Gram

Technical details

Model type

MultiModal

Train dataset

DataComp

Test dataset

DataComp

Hardware

GPU

CPU

Intel®

Core™ Ultra 5 125H (GPU)

Size (GB)

Original

1.71

CLIKA

0.55

68%

smaller

Accuracy (zero-shot top1)

Original

78.86

CLIKA

78.16

-0.7%

accuracy loss

Speed (FPS)

Original

1.24

CLIKA

7.82

6.3x

faster