Rems 150824 by Solista...
Posted: Sat Aug 17, 2024 9:23 am
Rems Engine by Solista
Unread post by Eduard Nemeth » Thu Aug 15, 2024 1:30 pm
========================================================
* Rems 150824 by Solista Rems is a small river on the banks of which I live.
The engine is based on SF dev and has an ultra-fast but excellent search. The latest SF NNUE networks are used, which can be downloaded and implemented here, currently "nn-9feb66a07029.nnue" is used.
If you have a fast computer, consider the multi-variant mode.
In the analysis I prefer to use 2 variants and have optimized this engine for it.
Features: Minimum Thinking Time, Slow Mover, external networks.
Download Win 64-Bit: https://pixeldrain.com/u/txXtCQuv
* Rems MPV 150824 by Solista. Version with Random Op. MultiPV mode.
The engine is based on SF dev and has an ultra-fast but still excellent search. The latest SF NNUE networks are used, currently "nn-9feb66a07029.nnue" is used.
Features: Minimum Thinking Time, Slow Mover, external networks.
2 variants are preset. The number of variants can be increased or reduced to 1. Download Win 64-bit: https://pixeldrain.com/u/WwmPru4S
* Rems EXP 150824 by Solista, this engine is based on a HypnoS dev version by Marco Zerbinati. However, I have implemented the rems-specific source code that distinguishes the engine from HypnoS.
All features of HypnoS can be used, including the learning file (Eman version). Any Eman.exp can be used, but it must be renamed as "Rems.exp".
I have been using this engine successfully for several weeks in practice on Lichess. It should be noted that with the BotLi tool, important features must be entered in this way in the BotLi yml file under Options:
MoveOverhead: 500
Slow mover: 120
Experience Enabled: true
Otherwise, the game is played with default setting, which means that the learning file is not activated and "Move Overhead" is not changed.
This is because the tool BotLi is oriented to Stockfish, where "Move Overhead" is written like that, but in HypnoS it’s like this: "MoveOverhead".
Download Win 64-Bit: https://pixeldrain.com/u/xzzPcSmq
* August 16, 2024: The test engine plays well on PlayChess.com, some players are just lying in the sun and not playing, they want to boast. Detlef Uter plays nonstop and against all opponents --> Picture, the balance of +8 on this even server is good.
My own tests on Lichess also went well, so here is this new version:
Rems EXP 160824: https://pixeldrain.com/u/ftge4tKV
The idea of creating a version with the Random Op MultiPV mode is very interesting to me. Such an engine would be tactically better and the saved learning values could later benefit the single-mode engine.
By the way: Analyses in normal multi-variant mode with 2 variants were excellent.
Here is the version with Random Op MultiPV mode, what’s the point ?
During a continuous analysis and also in the game, ratings are stored in the EXP file if this option is enabled.
This version is tactically stronger than the normal one and the tactical positions are better rated.
This benefits the normal version when it accesses the learning file.
Preset 2 variants, you can increase or decrease them to 1.
Download Rems EXT 160824: https://pixeldrain.com/u/N8hykJAH
For a better understanding and distinction of the engines, please read:
Rems 150824: The standard version with normal mode, new SF patches and all the latest NNUE networks are used.
* Rems EXT 160824: EXT stands for extended (extended), this version is identical to the Rems EXP version, but it will also be the Random Op. MultiPV mode support. This version has most features, see Engine options.
* August 17, 2024: Tip for using engines with the random op MultiPV mode. My practice has shown that the setting with 2 variants and plies = 6 (instead of 16) is better. There are only a few positions where more plies will do more, and the same is true for the number of variants.
With the suggested values, however, the search speed increases enormously.Test it yourself by changing the setting in the engine options.
I will probably set this setting as the default in future.
Thank you very much Eduard !
Unread post by Eduard Nemeth » Thu Aug 15, 2024 1:30 pm
========================================================
* Rems 150824 by Solista Rems is a small river on the banks of which I live.
The engine is based on SF dev and has an ultra-fast but excellent search. The latest SF NNUE networks are used, which can be downloaded and implemented here, currently "nn-9feb66a07029.nnue" is used.
If you have a fast computer, consider the multi-variant mode.
In the analysis I prefer to use 2 variants and have optimized this engine for it.
Features: Minimum Thinking Time, Slow Mover, external networks.
Download Win 64-Bit: https://pixeldrain.com/u/txXtCQuv
* Rems MPV 150824 by Solista. Version with Random Op. MultiPV mode.
The engine is based on SF dev and has an ultra-fast but still excellent search. The latest SF NNUE networks are used, currently "nn-9feb66a07029.nnue" is used.
Features: Minimum Thinking Time, Slow Mover, external networks.
2 variants are preset. The number of variants can be increased or reduced to 1. Download Win 64-bit: https://pixeldrain.com/u/WwmPru4S
* Rems EXP 150824 by Solista, this engine is based on a HypnoS dev version by Marco Zerbinati. However, I have implemented the rems-specific source code that distinguishes the engine from HypnoS.
All features of HypnoS can be used, including the learning file (Eman version). Any Eman.exp can be used, but it must be renamed as "Rems.exp".
I have been using this engine successfully for several weeks in practice on Lichess. It should be noted that with the BotLi tool, important features must be entered in this way in the BotLi yml file under Options:
MoveOverhead: 500
Slow mover: 120
Experience Enabled: true
Otherwise, the game is played with default setting, which means that the learning file is not activated and "Move Overhead" is not changed.
This is because the tool BotLi is oriented to Stockfish, where "Move Overhead" is written like that, but in HypnoS it’s like this: "MoveOverhead".
Download Win 64-Bit: https://pixeldrain.com/u/xzzPcSmq
* August 16, 2024: The test engine plays well on PlayChess.com, some players are just lying in the sun and not playing, they want to boast. Detlef Uter plays nonstop and against all opponents --> Picture, the balance of +8 on this even server is good.
My own tests on Lichess also went well, so here is this new version:
Rems EXP 160824: https://pixeldrain.com/u/ftge4tKV
The idea of creating a version with the Random Op MultiPV mode is very interesting to me. Such an engine would be tactically better and the saved learning values could later benefit the single-mode engine.
By the way: Analyses in normal multi-variant mode with 2 variants were excellent.
Here is the version with Random Op MultiPV mode, what’s the point ?
During a continuous analysis and also in the game, ratings are stored in the EXP file if this option is enabled.
This version is tactically stronger than the normal one and the tactical positions are better rated.
This benefits the normal version when it accesses the learning file.
Preset 2 variants, you can increase or decrease them to 1.
Download Rems EXT 160824: https://pixeldrain.com/u/N8hykJAH
For a better understanding and distinction of the engines, please read:
Rems 150824: The standard version with normal mode, new SF patches and all the latest NNUE networks are used.
* Rems EXT 160824: EXT stands for extended (extended), this version is identical to the Rems EXP version, but it will also be the Random Op. MultiPV mode support. This version has most features, see Engine options.
* August 17, 2024: Tip for using engines with the random op MultiPV mode. My practice has shown that the setting with 2 variants and plies = 6 (instead of 16) is better. There are only a few positions where more plies will do more, and the same is true for the number of variants.
With the suggested values, however, the search speed increases enormously.Test it yourself by changing the setting in the engine options.
I will probably set this setting as the default in future.
Thank you very much Eduard !
