Skip to content
Branch: master
Find file History
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
..
Failed to load latest commit information.
README.md
cifar_int8_cache
inference_aibox.py
resnet_cifar.uff

README.md

dawnbench_inference_imagenet

run inference task on ImageNet

The following instructions show how to achieve the performance that we submitted to DAWNBench step by step.

  1. install CUDA 10 and CUDNN 7, TensorRT 6 and TensorFlow 1.13
    download CUDA 10.0.130 for Ubuntu 16.04  (https://developer.nvidia.com/compute/cuda/10.0/Prod/local_installers/cuda-repo-ubuntu1604-10-0-local-10.0.130-410.48_1.0-1_amd64)
    download and install CUDNN 7.6.3.30 for Ubuntu 16.04 (https://developer.nvidia.com/compute/machine-learning/cudnn/secure/7.6.5.32/Production/10.0_20191031/Ubuntu16_04-x64/libcudnn7_7.6.5.32-1%2Bcuda10.0_amd64.deb)
    download and install TensorRT 6.0.1.5 for Ubuntu 16.04 (https://developer.nvidia.com/compute/machine-learning/tensorrt/secure/6.0/GA_6.0.1.5/local_repos/nv-tensorrt-repo-ubuntu1604-cuda10.0-trt6.0.1.5-ga-20190913_1-1_amd64.deb)
  1. install dependencies
	opencv==3.4.3
	libjpeg-turbo==2.0.1
	python==3.5.2 
	numpy==1.17.3 
	tensorflow==1.13.1 
  1. Download ModelArts-AIBOX from https://modelarts-labs.obs.cn-north-1.myhuaweicloud.com/tools/modelarts_aibox-0.13.0-py3-none-any.whl.
  2. Run
	pip install modelarts_aibox-0.13.0-py3-none-any.whl
	export LD_LIBRARY_PATH=/usr/local/lib/python3.5/dist-packages/modelarts/aibox:/usr/local/lib/python3.5/dist-packages/modelarts/aibox/operator/:/usr/local/cuda/lib64/:${LD_LIBRARY_PATH}
	nvidia-smi -pm 1
  1. Clone this repo.
  2. Modify 'DATA_DIR, UFF_FILE, CALIB_FILE' in inference_aibox.py if necessary.
  3. Run python inference_aibox.py.
You can’t perform that action at this time.