Leelenstein Chess

Return to site

Leelenstein Chess

Leelenstein Chess Engine

There is no single 'best net' for Leela but there a few worth recommending for various purposes. The most important consideration in choosing a net is picking the right size for your hardware and time controls (TC). In general, if you have a weak GPU (or no GPU at all) and you want to only spend milliseconds per move, then you want a smaller net that evaluates positions more quickly, i.e. higher NPS (nodes per second). On the other hand, if you have an RTX card (or several) and you want to run an analysis on a single move for hours at a time, then the quality of the evaluation is more important than the speed and a larger (but slower) net is probably going to work best.
Chess engine: Leelenstein 925. November 30, 2019 Leelenstein is a neural-network engine based on Lc0. Leela sacks a piece in the King's Gambit Accepted to attack Leelenstein's king in the center of the board. Leela vs Leelenstein King's Gambit Accepted: King'. Leelenstein is a derivative of Leela chess Zero and uses mostly the same source code except its training includes some non zero knowledge. It appears that Neural network chess engines continue to maintain a slig. Grandmaster (GM) is a title awarded to chess players by the world chess organization FIDE. Apart from World Champion, Grandmaster is the highest title a chess player can attain. Once achieved, the title is generally held for life, though exceptionally it may be revoked for cheating. The abbreviation IGM for International Grandmaster is also. Allie is a chess engine heavily inspired by the seminal AlphaZero paper and the Lc0 project. How is she related to Leela? Like Leela, Allie is based off of the same concepts and algorithms that.
Size versus Recommended Purpose<10b: Recommended for sparring vs humans
10b: Recommended for running on CPU
20b: Recommended for running on non-RTX cards or TC on the order of seconds (with RTX)
24b: Recommended for TC > 1 minute per move with an RTX card
>24b: Recommended for long analysis or when speed isn't a major factor
Network ListFor nets of the same size, the first net listed is likely the strongest.
SizeNameSource for DownloadNotes
5b x 48fGood Gyal 5GitHub: dkappe Bad GyalOther sizes also here
9b x 112fID11258-112x9-seGitHub: dkappe Distilled NetworksOther sizes also here
10b x 128fLatest T59lczero.org run 2 networks 591226.A previous test run
10b x 128fLittle Demon 2data.lczero.org repository (LD2)JH nets also here
16b x 192fJ20-460GitHub: jhorthos Leela TrainingTrained on T40 data
20b x 256fLeelenstein 14.2Patreon: jjoshEarly Access
20b x 256fSV-20b-t40-1541Sergio-V repositoryTrained on T40 data
20b x 256fLeelenstein 14.014.0 PostNo account required
20b x 256fT40: #42850lczero.org run 1 networksLast T40 net
24b x 320fLatest T60lczero.org run 1 networksCurrent main run
24b x 320fJ13B.2-136GitHub: jhorthos Leela Training'Terminator 2' Net
30b x 384f384x30-t60-3010, 384x30-t60-3044Sergio-V repositoryTrained on T60 data
30b x 384fSV-30b-t40-1705Sergio-V repositoryTrained on T40 data

Size	Name	Source for Download	Notes
5b x 48f	Good Gyal 5	GitHub: dkappe Bad Gyal	Other sizes also here
9b x 112f	ID11258-112x9-se	GitHub: dkappe Distilled Networks	Other sizes also here
10b x 128f	Latest T59	lczero.org run 2 networks 591226.	A previous test run
10b x 128f	Little Demon 2	data.lczero.org repository (LD2)	JH nets also here
16b x 192f	J20-460	GitHub: jhorthos Leela Training	Trained on T40 data
20b x 256f	Leelenstein 14.2	Patreon: jjosh	Early Access
20b x 256f	SV-20b-t40-1541	Sergio-V repository	Trained on T40 data
20b x 256f	Leelenstein 14.0	14.0 Post	No account required
20b x 256f	T40: #42850	lczero.org run 1 networks	Last T40 net
24b x 320f	Latest T60	lczero.org run 1 networks	Current main run
24b x 320f	J13B.2-136	GitHub: jhorthos Leela Training	'Terminator 2' Net
30b x 384f	384x30-t60-3010, 384x30-t60-3044	Sergio-V repository	Trained on T60 data
30b x 384f	SV-30b-t40-1705	Sergio-V repository	Trained on T40 data

Note: The Sergio-V nets are also available on data.lczero.org in some cases.
'This is all too complicated. Just tell me what net to use!'

There is no single 'best net' for Leela but there a few worth recommending for various purposes. The most important consideration in choosing a net is picking the right size for your hardware and time controls (TC). In general, if you have a weak GPU (or no GPU at all) and you want to only spend milliseconds per move, then you want a smaller net that evaluates positions more quickly, i.e. higher NPS (nodes per second). On the other hand, if you have an RTX card (or several) and you want to run an analysis on a single move for hours at a time, then the quality of the evaluation is more important than the speed and a larger (but slower) net is probably going to work best.
Chess engine: Leelenstein 925. November 30, 2019 Leelenstein is a neural-network engine based on Lc0. Leela sacks a piece in the King's Gambit Accepted to attack Leelenstein's king in the center of the board. Leela vs Leelenstein King's Gambit Accepted: King'. Leelenstein is a derivative of Leela chess Zero and uses mostly the same source code except its training includes some non zero knowledge. It appears that Neural network chess engines continue to maintain a slig. Grandmaster (GM) is a title awarded to chess players by the world chess organization FIDE. Apart from World Champion, Grandmaster is the highest title a chess player can attain. Once achieved, the title is generally held for life, though exceptionally it may be revoked for cheating. The abbreviation IGM for International Grandmaster is also. Allie is a chess engine heavily inspired by the seminal AlphaZero paper and the Lc0 project. How is she related to Leela? Like Leela, Allie is based off of the same concepts and algorithms that.
Size versus Recommended Purpose<10b: Recommended for sparring vs humans
10b: Recommended for running on CPU
20b: Recommended for running on non-RTX cards or TC on the order of seconds (with RTX)
24b: Recommended for TC > 1 minute per move with an RTX card
>24b: Recommended for long analysis or when speed isn't a major factor
Network ListFor nets of the same size, the first net listed is likely the strongest.
SizeNameSource for DownloadNotes
5b x 48fGood Gyal 5GitHub: dkappe Bad GyalOther sizes also here
9b x 112fID11258-112x9-seGitHub: dkappe Distilled NetworksOther sizes also here
10b x 128fLatest T59lczero.org run 2 networks 591226.A previous test run
10b x 128fLittle Demon 2data.lczero.org repository (LD2)JH nets also here
16b x 192fJ20-460GitHub: jhorthos Leela TrainingTrained on T40 data
20b x 256fLeelenstein 14.2Patreon: jjoshEarly Access
20b x 256fSV-20b-t40-1541Sergio-V repositoryTrained on T40 data
20b x 256fLeelenstein 14.014.0 PostNo account required
20b x 256fT40: #42850lczero.org run 1 networksLast T40 net
24b x 320fLatest T60lczero.org run 1 networksCurrent main run
24b x 320fJ13B.2-136GitHub: jhorthos Leela Training'Terminator 2' Net
30b x 384f384x30-t60-3010, 384x30-t60-3044Sergio-V repositoryTrained on T60 data
30b x 384fSV-30b-t40-1705Sergio-V repositoryTrained on T40 data
Note: The Sergio-V nets are also available on data.lczero.org in some cases.
'This is all too complicated. Just tell me what net to use!'If you don't care about squeezing out the very best performance for a particular situation and want a general-purpose net, pick a medium size 20b net, which should do reasonably well (if not optimal) under most common conditions.
As in the table above, the top-recommended 20b net is 256x20-t40-1541.pb.gz from Sergio Vieri's repository.
Leelenstein Chess EngineIf this page hasn't been updated recently, check the Discord channels for a recommendation. Be sure to specify your hardware and use case so the helpful regulars know what to recommend.