SF-NN embedded

Discussion of anything and everything relating to chess playing software and machines.

Moderators: Harvey Williamson, Dann Corbit, hgm

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
corres
Posts: 3600
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: SF-NN embedded

Post by corres » Fri Aug 28, 2020 6:40 am

I think for the stable working it needs stable connection between Stockfish and its NN, it is obvious.
I could reach this aim with using generalized NN name (nn.bin) and the simplified code (Stockfish can not show the name of NN). Making difference among net its date is a good tool when I change the NN.
Embedding the NN of SF+NNUE is a good thought if we do not want to change the used NN and Sergio use only the specified (just developed Stockfish) for making the NN specified to the just developed Stockfish. If he used positions gained from different games there is no any sense embedding the NN.

muppetmuppet
Posts: 8
Joined: Tue May 19, 2020 5:41 pm
Full name: Ian Hodges

Re: SF-NN embedded

Post by muppetmuppet » Fri Aug 28, 2020 3:37 pm

In the latest r5 release tablebases don't seem to work for honey or bluefish. I've never had a problem before with these and my other engines are still fine but the new engines fail to find them even though the path is specified. The path is super short too so its not a long path problem which allie used to have.

Anyone else have this problem or is there something obvious I am missing ?

User avatar
MikeB
Posts: 4488
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: SF-NN embedded

Post by MikeB » Fri Aug 28, 2020 4:08 pm

muppetmuppet wrote:
Fri Aug 28, 2020 3:37 pm
In the latest r5 release tablebases don't seem to work for honey or bluefish. I've never had a problem before with these and my other engines are still fine but the new engines fail to find them even though the path is specified. The path is super short too so its not a long path problem which allie used to have.

Anyone else have this problem or is there something obvious I am missing ?
not here - OS and GUI? and can you provide the path you are using?
Image

muppetmuppet
Posts: 8
Joined: Tue May 19, 2020 5:41 pm
Full name: Ian Hodges

Re: SF-NN embedded

Post by muppetmuppet » Fri Aug 28, 2020 8:33 pm

windows 10. arena3.51. c:\TB. TB are on an ssd in case that matters. Arena is on HDD so is the engine folder.

User avatar
MikeB
Posts: 4488
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: SF-NN embedded

Post by MikeB » Sat Aug 29, 2020 4:02 am

muppetmuppet wrote:
Fri Aug 28, 2020 8:33 pm
windows 10. arena3.51. c:\TB. TB are on an ssd in case that matters. Arena is on HDD so is the engine folder.
I do not use Arena as it is inherently buggy for me. Try using BanksiaGui, it has a nice engine log that shows you what is happening and one can see syzygy being loaded.



Oki Maguro XIr5 < setoption name SyzygyPath value c:/syzygy
Oki Maguro XIr5 < isready
Oki Maguro XIr5 > info string Hash Table: Windows Large Pages, 256 Mb
Oki Maguro XIr5 > info string Hash Table: Windows Large Pages, 20480 Mb
Oki Maguro XIr5 < ucinewgame
Oki Maguro XIr5 < position fen 8/p4p1Q/8/p7/p6p/K7/2r5/1k6 w - -
Oki Maguro XIr5 < go infinite
Oki Maguro XIr5 > info string Found 510 tablebases
Oki Maguro XIr5 > readyok
Oki Maguro XIr5 > info string Found 510 tablebases


for reasons I can't explain, most of the GUIs I now use in Windows take the unix looking forward slash and not the typical backslash one would see in Windows - that could be the issue in Arena - but again I don't use Arena and not I will not be using it, I now use cutechess-cli for matches and BanksiaGUI for analysis. LucasChess and WinBoard are good for playing engines, along with BanksiaGUI.
Image

Dann Corbit
Posts: 11702
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: SF-NN embedded

Post by Dann Corbit » Sat Aug 29, 2020 4:24 am

I think the embedded file is misguided.
For instance, I build a bunch of different binaries, and shall each one have a 20MB pile of goo sitting in it's data space?
Imagine if all of the stockfish variants follow suit.
Then we will have 3-4x 20MB pile multiplied by variant count.
I think it would make me cry,.

There is a reason we don't embed 5 man tablebase file sets in the binaries.
If it were that hard to browse to a file, nobody could use a tablebase.

Please, for the love of all that is good and true, no more embedding of 20MB files.

I move files across the network like mad. The volume of the file makes the virus scan time go up something like quadratically. A 20 MB file causes a big delay. A 60 MB file causes an atrocious delay. I do a build for each new variant of each new engine. We are talking about serious problems here.

Please, please, please. I am begging down on my knees. No embedding of TB files or Networks.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

Michel
Posts: 2110
Joined: Sun Sep 28, 2008 11:50 pm

Re: SF-NN embedded

Post by Michel » Sat Aug 29, 2020 8:15 am

Dann Corbit wrote:
Sat Aug 29, 2020 4:24 am
I think the embedded file is misguided.
For instance, I build a bunch of different binaries, and shall each one have a 20MB pile of goo sitting in it's data space?
Imagine if all of the stockfish variants follow suit.
Then we will have 3-4x 20MB pile multiplied by variant count.
I think it would make me cry,.

There is a reason we don't embed 5 man tablebase file sets in the binaries.
If it were that hard to browse to a file, nobody could use a tablebase.

Please, for the love of all that is good and true, no more embedding of 20MB files.

I move files across the network like mad. The volume of the file makes the virus scan time go up something like quadratically. A 20 MB file causes a big delay. A 60 MB file causes an atrocious delay. I do a build for each new variant of each new engine. We are talking about serious problems here.

Please, please, please. I am begging down on my knees. No embedding of TB files or Networks.
Embedding is optional. Distros will very likely distribute SF without an embedded net.
Ideas=science. Simplification=engineering.
Without ideas there is nothing to simplify.

muppetmuppet
Posts: 8
Joined: Tue May 19, 2020 5:41 pm
Full name: Ian Hodges

Re: SF-NN embedded

Post by muppetmuppet » Sat Aug 29, 2020 9:40 am

amusingly honey is winning my current bullet top 20 engines tournament with no TB's. Maybe they are actually bad at very fast TC.

Joerg Oster
Posts: 791
Joined: Fri Mar 10, 2006 3:29 pm
Location: Germany

Re: SF-NN embedded

Post by Joerg Oster » Sat Aug 29, 2020 10:07 am

Michel wrote:
Sat Aug 29, 2020 8:15 am
Dann Corbit wrote:
Sat Aug 29, 2020 4:24 am
I think the embedded file is misguided.
For instance, I build a bunch of different binaries, and shall each one have a 20MB pile of goo sitting in it's data space?
Imagine if all of the stockfish variants follow suit.
Then we will have 3-4x 20MB pile multiplied by variant count.
I think it would make me cry,.

There is a reason we don't embed 5 man tablebase file sets in the binaries.
If it were that hard to browse to a file, nobody could use a tablebase.

Please, for the love of all that is good and true, no more embedding of 20MB files.

I move files across the network like mad. The volume of the file makes the virus scan time go up something like quadratically. A 20 MB file causes a big delay. A 60 MB file causes an atrocious delay. I do a build for each new variant of each new engine. We are talking about serious problems here.

Please, please, please. I am begging down on my knees. No embedding of TB files or Networks.
Embedding is optional. Distros will very likely distribute SF without an embedded net.
It's just that almost nobody is interested or relying on what distro packagers do ...

Some compile from the sources, but the vast majority simply take the executable from abrok or stockfishchess.org,
or those provided by the community.
Jörg Oster

User avatar
MikeB
Posts: 4488
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: SF-NN embedded

Post by MikeB » Sat Aug 29, 2020 4:12 pm

Dann Corbit wrote:
Sat Aug 29, 2020 4:24 am
I think the embedded file is misguided.
For instance, I build a bunch of different binaries, and shall each one have a 20MB pile of goo sitting in it's data space?
Imagine if all of the stockfish variants follow suit.
Then we will have 3-4x 20MB pile multiplied by variant count.
I think it would make me cry,.

There is a reason we don't embed 5 man tablebase file sets in the binaries.
If it were that hard to browse to a file, nobody could use a tablebase.

Please, for the love of all that is good and true, no more embedding of 20MB files.

I move files across the network like mad. The volume of the file makes the virus scan time go up something like quadratically. A 20 MB file causes a big delay. A 60 MB file causes an atrocious delay. I do a build for each new variant of each new engine. We are talking about serious problems here.

Please, please, please. I am begging down on my knees. No embedding of TB files or Networks.
I see this as a specific usage case, and your usage case would not qualify. More for the uninitiated in the finer details of engine management ,i.e., the casual user.
Image

Post Reply