cbench 1.3.1

A cross-platform client to perform collaborative and reproducible benchmarking, optimization and co-design of software and hardware for emerging workloads (AI, ML, quantum, IoT) via the open cKnowledge.io portal

Linux/MacOS: Windows:

News

We have successfully completed the prototyping phase of the Collective Knowledge technology to make it easier to reproduce AI&ML and deploy it in production with the help of portable CK workflows, reusable artifacts and MLOps as described in this white paper and the CK presentation. We are now preparing the second phase of this project to make CK simpler to use, more stable and more user friendly - don't hesitate to get in touch with the CK author to know more!

Introduction

cBench is a small and cross-platform framework connected with the open Collective Knowledge portal to help researchers and practitioners reproduce ML&systems research on their own bare-metal platforms, participate in collaborative benchmarking and optimization, and share results on live scoreobards.

You can try to reproduce MLPerf inference benchmark on your machine using this solution and see public results from the community on this scoreboard.

cBench is a part of the Collective Knowledge project (CK) and uses portable CK solutions to describe how to download, build, benchmark and optimize applications across different hardware, software, models and data sets.

Platform support:

	As a host platform	As a target platform
Generic Linux	✓	✓
Linux (Arm)	✓	✓
Raspberry Pi	✓	✓
MacOS	✓	±
Windows	✓	✓
Android	±	✓
iOS	TBD	TBD

Object detection crowd-benchmarking demo on Ubuntu

Install prerequisites:

sudo apt update
sudo apt install git wget libz-dev curl cmake
sudo apt install gcc g++ autoconf autogen libtool
sudo apt install libfreetype6-dev
sudo apt install python3.7-dev
sudo apt install -y libsm6 libxext6 libxrender-dev

Install cbench:

python3 -m pip install cbench

Initialize the CK solution for MLPerf:

cb init demo-obj-detection-coco-tf-cpu-benchmark-linux-portable-workflows

Participate in crowd-benchmarking:

cb benchmark demo-obj-detection-coco-tf-cpu-benchmark-linux-portable-workflows

See your results on a public SOTA dashboard.

You can also use the stable Docker image to participate in crowd-benchmarking:

sudo docker run ctuning/cbench-obj-detection-coco-tf-cpu-benchmark-linux-portable-workflows /bin/bash -c "cb benchmark demo-obj-detection-coco-tf-cpu-benchmark-linux-portable-workflows"

You can also check all dependencies for this solution.

Documentation

• Online docs for the Collective Knowledge technology

Feedback

• This is an ongoing project - don't hesitate to contact us if you have any feedback and suggestions!

Acknowledgments

We would like to thank all CK partners for fruitful discussions and feedback!

Copyright 2020 cTuning foundation