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