GithubHelp home page GithubHelp logo

jwatzman / openbench Goto Github PK

View Code? Open in Web Editor NEW

This project forked from andygrant/openbench

0.0 1.0 0.0 46.47 MB

OpenBench is a Distributed SPRT Testing Framework for Chess Engines

License: GNU General Public License v3.0

Shell 0.22% Python 75.64% CSS 2.60% HTML 21.55%

openbench's Introduction

OpenBench

OpenBench is an open-source Chess Engine Testing Framework for UCI engines. OpenBench provides a lightweight interface and client to facilitate running fixed-game tests as well as SPRT tests to benchmark changes to engines for performance and stability. OpenBench supports Fischer Random Chess.

OpenBench is the primary testing framework used for the development of Ethereal. The primary instance of OpenBench can be found at http://chess.grantnet.us. The Primary instance of OpenBench supports development for Berserk, Bit-Genie, BlackMarlin, Demolito, Drofa, Ethereal, FabChess, Halogen, Igel, Koivisto, Laser, RubiChess, Seer, Stash, Weiss, Winter, and Zahak. A dozen or more engines are using their own private, local instances of OpenBench.

You can join OpenBench's Discord server to join the discussion, see what developers are working on and talking about, or to find out how you can contribute to the project and become a part of it. OpenBench is heavily inspired by Fishtest. The project is powered by the Django Web Framework and Cutechess.

The Client

The OpenBench Client is a python3 script. The Client needs access to make and gcc. Make is used to initiate builds for engines. Every engine on the framework will have a makefile in its repository. That makefile will execute a compiler, which is configured via OpenBench. When the Client is run, it will first check for compilers against the list of compilers requested by the engines on the framework. gcc is needed though, even if no C engines are on the framework, in order to determine CPU flags like POPCNT, AVX, AVX2, and more. For Windows users, a POSIX version of gcc is recommended.

The client takes four arguments: Username, Password, Server, Threads. First create an account via the instance's webpage. The instance will also be the Server provided to the client. Threads tells the Client how many games can be run in parallel. This should be no more than your CPU count. An example: python3 OpenBench.py -U username -P password -S http://chess.grantnet.us/ -T 8

The client will create an Engines, Books, Networks, and PGNs directory. These are used to store compiled engines, downloaded opening books, downloaded Networks, and PGNs of the games played. By default, Engines and PGNs are deleted after 24 hrs. Networks are deleted after a month. These durations can be changed directly in the Client in cleanup_client()

Engine Compliance with OpenBench

In order for many engines to operate under a shared framework, each engine must have uniform compliance in a small number of aspects. The following three paragraphs outline the standards expected by OpenBench.

Every test on OpenBench must have a Threads= and Hash= set in the UCI options. This is because the thread count plays a role in determining how many games to run in parallel. The Hash should, by default, be set to a low value. This is because when preparing an engine on the Client, benchmarks are run for each thread requested by the Client. A machine with 32 threads will run 32 copies of the engine at the same time during benching.

The engine must support being run from the command line with a sole argument, bench. This should execute a series of low-depth searches on a set of positions. The resulting searches should be summed up, and a final Node count and Nodes Per Second value should be printed. In Ethereal, the final line of output before exiting contains 3938740 nodes 1992281 nps. The Node value is needed as a sort of checksum for the engine. The NPS value is needed in order to scale machines of differing speeds to one uniform time control.

The engine must have a makefile, with a default target that builds a single binary, whose name is determined by the EXE= argument. If the engine may be compiled with multiple compilers (such as supporting gcc & clang at the same time), then the CC= argument must also be accepted. Finally, engines with NNUE files must allow building via EVALFILE=, which will compile the Network weights into the binary.

Lastly, an engine should play nice when closed by error, by Cutechess, by python, or by any other means. This can be verified by checking for hanging processes. Generally this is not an issue, but poor code for reading input pipes can cause this to occur. This is crucial, as the majority of machines connected to the main OpenBench instance are not constantly monitored.

Adding an Engine to OpenBench

To add an engine to an existing OpenBench framework, all that must be done is to include an additional entry in the OPENBENCH_CONFIG dictionary located in OpenBench/config.py. The entry for Ethereal looks like the following:

'Ethereal' : {

    'nps'    : 1200000,
    'base'   : 'master',
    'book'   : 'Pohl.epd',
    'bounds' : '[0.00, 5.00]',
    'source' : 'https://github.com/AndyGrant/Ethereal',

    'build' : {
        'path'      : 'src',
        'compilers' : ['gcc'],
        'cpuflags'  : ['AVX2', 'AVX', 'FMA', 'POPCNT', 'SSE2', 'SSE'],
    },

    'testmodes' : {
        'stc'     : { 'threads' : 1, 'hash' :   8, 'timecontrol' : '10.0+0.1' },
        'ltc'     : { 'threads' : 1, 'hash' :  64, 'timecontrol' : '60.0+0.6' },
        'smpstc'  : { 'threads' : 8, 'hash' :  64, 'timecontrol' : '5.0+0.05' },
        'smpltc'  : { 'threads' : 8, 'hash' : 256, 'timecontrol' : '20.0+0.2' },
     },
},

The nps field is the speed of a single Ethereal process, when a copy is run on each thread of a machine at the same time. The value is subjective, and acts only to scale different machines to a uniform speed. The main OpenBench instance is scaled to the 16 threads of a Ryzen 3700x. Scripts/bench_engine.py allows you to simulate this process.

The base field refers to the base branch for testing. Generally, one will test against master. New repositories default to main instead of master, because of wokeness. This field is simply the default auto-filled value for creating a new test. It can be changed at any time.

The book field refers to the default opening book to be used. By default, OpenBench supports many different opening books, including fischer books, and double-fischer books.

The bounds field refers to the default SPRT bounds when creating a test. The first value is elo0, and the second value is elo1. Again, this is used to auto-fill test creation fields. It can be changed at any time.

The source field refers to the location of the repository that is being used. This is only used to set the links in the sidebar. Individual users set their own repositories for auto-filling test creation fields.

The path field refers to the location of the Makefile in the engine's repository. No leading or trailing slashes should be included.

The compilers field allows a list of compilers that are able to build the engine. Version requirements may be set as well. For example, ['gcc>=8.0.0'] would require the Client to have a gcc version at or above v8.0.0.

The cpuflags field allows a list of required CPU flags. Generally, this is not needed. Ethereal has these set just to make sure that all machines are able to run the NNUE at the fastest speeds. The option may be left as an empty list.

openbench's People

Contributors

andygrant avatar jtseng20 avatar jwatzman avatar mhouppin avatar ppigazzini avatar terjekir avatar theo77186 avatar vshcherbyna avatar

Watchers

 avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    ๐Ÿ–– Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. ๐Ÿ“Š๐Ÿ“ˆ๐ŸŽ‰

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google โค๏ธ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.