Difference between revisions of "A0lite"

From Chessprogramming wiki
Jump to: navigation, search
 
Line 4: Line 4:
 
a didactic [[UCI]] compliant [[Neural Networks|neural network]] chess engine by [[Dietrich Kappe]], written in [[Python]], released in March 2020 under the permissive [[Massachusetts Institute of Technology#License|MIT License]] <ref>[http://www.talkchess.com/forum3/viewtopic.php?f=2&t=73495 New engine: a0lite] by [[Dietrich Kappe]], [[CCC]], March 28, 2020</ref> as successor of '''LeelaLite''', already announced in October 2018 <ref>[http://www.talkchess.com/forum3/viewtopic.php?f=2&t=68789 Leela Lite: A toolkit for experimenting with leela nets in python] by [[Dietrich Kappe]], [[CCC]],  October 31, 2018</ref>.
 
a didactic [[UCI]] compliant [[Neural Networks|neural network]] chess engine by [[Dietrich Kappe]], written in [[Python]], released in March 2020 under the permissive [[Massachusetts Institute of Technology#License|MIT License]] <ref>[http://www.talkchess.com/forum3/viewtopic.php?f=2&t=73495 New engine: a0lite] by [[Dietrich Kappe]], [[CCC]], March 28, 2020</ref> as successor of '''LeelaLite''', already announced in October 2018 <ref>[http://www.talkchess.com/forum3/viewtopic.php?f=2&t=68789 Leela Lite: A toolkit for experimenting with leela nets in python] by [[Dietrich Kappe]], [[CCC]],  October 31, 2018</ref>.
 
A0lite applies [[UCT|upper confidence bounds]] to [[Monte-Carlo Tree Search|Monte-Carlo trees]], and requires the installation of the [[Bad Gyal]] [https://en.wikipedia.org/wiki/PyTorch PyTorch] net evaluator, by default using ''MeanGirl-8 (32x4)'' net on CPU <ref>[https://github.com/dkappe/a0lite/blob/master/README.md a0lite/README.md at master · dkappe/a0lite · GitHub]</ref>. A0lite had its official tournament debut at the [[TCEC Season 19#Fourth|Qualification League]] of [[TCEC Season 19]].  
 
A0lite applies [[UCT|upper confidence bounds]] to [[Monte-Carlo Tree Search|Monte-Carlo trees]], and requires the installation of the [[Bad Gyal]] [https://en.wikipedia.org/wiki/PyTorch PyTorch] net evaluator, by default using ''MeanGirl-8 (32x4)'' net on CPU <ref>[https://github.com/dkappe/a0lite/blob/master/README.md a0lite/README.md at master · dkappe/a0lite · GitHub]</ref>. A0lite had its official tournament debut at the [[TCEC Season 19#Fourth|Qualification League]] of [[TCEC Season 19]].  
 +
 +
=Quotes=
 +
[[Dietrich Kappe]] explained his motivation for writing A0lite on [[CCC]], Mar 06, 2021 <ref>[http://talkchess.com/forum3/viewtopic.php?f=2&t=76787 Re: Your motivation for writing a chess engine?] by [[Dietrich Kappe]], Mar 06, 2021</ref> :
 +
 +
1. Teaching other people how simple it is to write a basic mcts/nn engine with a0lite python.
 +
2. Experimenting with new nn architectures and non-RL training approaches.
 +
3. Combining ab/nnue and mcts/nn in a hybrid approach. (Was a0lite julia, renamed Bender)
 +
4. Reach 3300 ccrl
 +
  
 
=See also=
 
=See also=

Latest revision as of 03:08, 16 August 2022

Home * Engines * A0lite

A0lite,
a didactic UCI compliant neural network chess engine by Dietrich Kappe, written in Python, released in March 2020 under the permissive MIT License [1] as successor of LeelaLite, already announced in October 2018 [2]. A0lite applies upper confidence bounds to Monte-Carlo trees, and requires the installation of the Bad Gyal PyTorch net evaluator, by default using MeanGirl-8 (32x4) net on CPU [3]. A0lite had its official tournament debut at the Qualification League of TCEC Season 19.

Quotes

Dietrich Kappe explained his motivation for writing A0lite on CCC, Mar 06, 2021 [4] :

1. Teaching other people how simple it is to write a basic mcts/nn engine with a0lite python.
2. Experimenting with new nn architectures and non-RL training approaches.
3. Combining ab/nnue and mcts/nn in a hybrid approach. (Was a0lite julia, renamed Bender)
4. Reach 3300 ccrl


See also

Forum Posts

External Links

Engine

Misc

References

Up one Level