Skip to content

Pytorch Classification quantsim models

Compare
Choose a tag to compare
@quic-bharathr quic-bharathr released this 26 Aug 23:05
· 18 commits to develop since this release

Optimized W4A8 and W8A8 checkpoints and encodings for Pytorch Torchvision instances of Resnet18, Resnet50 and Regnet_x_3_2gf.

  1. Cross-Layer Equalization and AdaRound have been applied in per channel mode on the original model for both INT4 and INT8 optimization
  2. "TF_enhanced" quantscheme was used for quantsim