python -m minimum.minimum_demo

# download imagenet dataset
cd QAT.axera
mkdir -p dataset/imagenet && cd dataset/imagenet
wget https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_train.tar --no-check-certificate
wget https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_val.tar --no-check-certificate
wget https://image-net.org/data/ILSVRC/2012/ILSVRC2012_devkit_t12.tar.gz --no-check-certificate

# download resnet50 pretrained model
cd QAT.axera
wget -O resnet50/resnet50_pretrained_float.pth https://download.pytorch.org/models/resnet50-0676ba61.pth

# train
cd QAT.axera
mkdir -p resnet50/checkpoint
python -m resnet50.train
# 4bit 量化位宽时参考 resnet50/config_4w4f 配置，并用 simplify_and_fix_4bit_dtype 替代 onnx_simplify

# test
cd QAT.axera
python -m resnet50.test