Fusions · Quantization · Compilation

Deploy PyTorch models
on edge hardware.

Hardware-aware graph transforms, fusions, and optimal quantization
in PyTorch.

by Embedl

Click a card to compare the original model with the Embedl quantized version

ResNet-50 CNN
25M params · 50 layers · 497 nodes
Original + Embedl Quantized
ViT-B/16 Transformer
86M params · 12 heads · 280 nodes
Original + Embedl Quantized
ConvNext Tiny ConvNet
28M params · 4 stages · 340 nodes
Original + Embedl Quantized
or try your own model

Drop your .pt2 file here

or