fx

Step-by-Step

This document describes the step-by-step instructions for reproducing PyTorch ResNest50 tuning and benchmarking results with Intel® Neural Compressor.

Prerequisite

1. Environment

Python 3.6 or higher version is recommended. The dependent packages are all in requirements, please install as following.

cd examples/pytorch/image_recognition/resnest/quantization/ptq/fx
pip install -r requirements.txt

2. Prepare Dataset

Download ImageNet Raw image to dir: /path/to/imagenet. The dir should include below folder:

ls /path/to/imagenet
train  val

3. Pytorch Models

GitHub Install

git clone https://github.com/zhanghang1989/ResNeSt.git
cd ResNeSt
git checkout 1dfb3e8867e2ece1c28a65c9db1cded2818a2031
python setup.py install
cd ..

Load Models

• Load using Torch Hub

import torch
# get list of models and save to cache
torch.hub.list('zhanghang1989/ResNeSt', force_reload=True)
# load pretrained models, using ResNeSt-50 as an example
net = torch.hub.load('zhanghang1989/ResNeSt', 'resnest50', pretrained=True)

or

• Load using python package

# using ResNeSt-50 as an example
from resnest.torch import resnest50
net = resnest50(pretrained=True)

Run

1. Quantization

python -u verify.py --tune --model resnest50 --batch-size 30 --workers 1 --no-cuda /path/to/imagenet

2. Benchmark

# int8
sh run_benchmark.sh --int8=true --mode=performance --input_model=resnest50  --dataset_location=/path/to/imagenet --config=saved_results
# fp32
sh run_benchmark.sh --mode=performance --input_model=resnest50  --dataset_location=/path/to/imagenet