Search
PKU Unveils Groundbreaking 2-bit Quantization Model for Efficient Inference on Mobile Devices
2025-08-16 /
Read about 0 minute
Author：小编
The inference process of large models demands substantial storage and computational resources, partly attributed to the extensive use of FP16 storage for weights. To address this, the PKU team has introduced the pioneering iFairy solution, which quantizes model weights to the complex number set {+1, -1, +i, -i}. These four values, each represented by 2 bits, achieve an impressive compression ratio of 1/8. This innovative approach is poised to dramatically reduce both storage and computational demands, enabling deployment on mobile phones with minimal overhead.
-
C114 Communication Network -
Communication Home
7 X 24 Track global technological trends
