Anyone else having issues getting today's Stockfish update to run?

[Leto]

I'm having an issue getting today's update of Stockfish to run in hybrid mode (NNUE enabled). I'm trying to get the one with the timestamp of 1597063097 to work and it's not working for me on either Aquarium or Fritz 17 GUI. I use the build intended for haswell CPU's running on windows, which has the .bmi at the end of the filename. On Fritz 17 GUI the engine loads but when I hit infinite analysis it gives me an error "Stockfish 100820" caused an exception. This is a problem in the engine."

However when I disable hybrid mode the engine runs fine on both GUI's. Is the network file nn-112bb1c8cdb5.nnue corrupted? The CRC-32 value I get for the network file is 3E7C97AE7F79E6D5, and the SHA-256 value is 112BB1C8CDB58B43324F765461DCA727BD468D610D257203E0D950515C15CFCA.

[zullil]

I'm having an issue getting today's update of Stockfish to run in hybrid mode (NNUE enabled). I'm trying to get the one with the timestamp of 1597063097 to work and it's not working for me on either Aquarium or Fritz 17 GUI. I use the build intended for haswell CPU's running on windows, which has the .bmi at the end of the filename. On Fritz 17 GUI the engine loads but when I hit infinite analysis it gives me an error "Stockfish 100820" caused an exception. This is a problem in the engine."

However when I disable hybrid mode the engine runs fine on both GUI's. Is the network file nn-112bb1c8cdb5.nnue corrupted? The CRC-32 value I get for the network file is 3E7C97AE7F79E6D5, and the SHA-256 value is 112BB1C8CDB58B43324F765461DCA727BD468D610D257203E0D950515C15CFCA.

i doubt the file's corrupted, since your SHA-256 matches the filename: 112BB1C8CDB5.

I can't even download the file; seems to be some problem at https://tests.stockfishchess.org/nns

[Leto]

I'm having an issue getting today's update of Stockfish to run in hybrid mode (NNUE enabled). I'm trying to get the one with the timestamp of 1597063097 to work and it's not working for me on either Aquarium or Fritz 17 GUI. I use the build intended for haswell CPU's running on windows, which has the .bmi at the end of the filename. On Fritz 17 GUI the engine loads but when I hit infinite analysis it gives me an error "Stockfish 100820" caused an exception. This is a problem in the engine."

However when I disable hybrid mode the engine runs fine on both GUI's. Is the network file nn-112bb1c8cdb5.nnue corrupted? The CRC-32 value I get for the network file is 3E7C97AE7F79E6D5, and the SHA-256 value is 112BB1C8CDB58B43324F765461DCA727BD468D610D257203E0D950515C15CFCA.

i doubt the file's corrupted, since your SHA-256 matches the filename: 112BB1C8CDB5.

I can't even download the file; seems to be some problem at https://tests.stockfishchess.org/nns

For some strange reason the hybrid mode works fine when I rename the network file to nn.bin, and change the evalfile path to ./eval/nn.bin

[MikeB]

I'm having an issue getting today's update of Stockfish to run in hybrid mode (NNUE enabled). I'm trying to get the one with the timestamp of 1597063097 to work and it's not working for me on either Aquarium or Fritz 17 GUI. I use the build intended for haswell CPU's running on windows, which has the .bmi at the end of the filename. On Fritz 17 GUI the engine loads but when I hit infinite analysis it gives me an error "Stockfish 100820" caused an exception. This is a problem in the engine."

However when I disable hybrid mode the engine runs fine on both GUI's. Is the network file nn-112bb1c8cdb5.nnue corrupted? The CRC-32 value I get for the network file is 3E7C97AE7F79E6D5, and the SHA-256 value is 112BB1C8CDB58B43324F765461DCA727BD468D610D257203E0D950515C15CFCA.

Just refreshed my fork last night - no issues here with Fritz 17 - but I notice some reports on Github:
https://github.com/official-stockfish/S ... ssues/2963

This is NOT an official clone as the bench signature is not aligned, but I don't think you will mind.... ;>)

Code: Select all

I
PGN File: c:/cluster.mfb/pgn/08100236.pgn
Time Control: Time Control-> base+inc: 50+1.000
Games: 20000
Threads: 1
Hash: 512

Current date : time (EDST)
Date: 08/10/20 : 10:16:43

Projected-> Time: 21h:50m:56s
     Run -> Time: 7h:40m:9s

7092 game(s) loaded
Rank Name            Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR
---------------------------------------------------------------------------------------------------------

   1 SF-XIr4-1351     3520   0.0    8    8  2836 1518.5  53.5  758  557 1521  26.7  53.6  3495
   2 Blue-XIr4-1351   3512   8.1    9    9  2838 1478.0  52.1  742  624 1472  26.1  51.9  3497
   3 stockfish-1351   3510   1.7    8    8  2838 1470.5  51.8  716  613 1509  25.2  53.2  3497
   4 Ho-XIr4-1351     3508   2.0    9    9  2836 1459.5  51.5  706  623 1507  24.9  53.1  3498
   5 BD-XIr4-1351     3450  58.3    8    8  2836 1165.5  41.1  436  941 1459  15.4  51.4  3513
---------------------------------------------------------------------------------------------------------

  Δ = delta from the next higher rated opponent
  # = number of games played
  Σ = total score, 1 point for win, 1/2 point for draw

LOS:
                SF Bl st Ho BD
SF-XIr4-1351       92 96 98100
Blue-XIr4-1351   7    62 74100
stockfish-1351   3 37    63100
Ho-XIr4-1351     1 25 36   100
BD-XIr4-1351     0  0  0  0

7092 game(s) loaded

loops/scheduled: 67/200

https://www.dropbox.com/t/lschiRUvc3Vw5WoM

To make this compliant with the license:
Source:
https://github.com/MichaelB7/Stockfish/tree/honey

[Jouni]

They update SF too fast now - up to 10 patches a day! Better to wait week or two and see how things settle. There has been a lot +10 ELO patches in addition to net improving.

[Leto]

I'm having an issue getting today's update of Stockfish to run in hybrid mode (NNUE enabled). I'm trying to get the one with the timestamp of 1597063097 to work and it's not working for me on either Aquarium or Fritz 17 GUI. I use the build intended for haswell CPU's running on windows, which has the .bmi at the end of the filename. On Fritz 17 GUI the engine loads but when I hit infinite analysis it gives me an error "Stockfish 100820" caused an exception. This is a problem in the engine."

However when I disable hybrid mode the engine runs fine on both GUI's. Is the network file nn-112bb1c8cdb5.nnue corrupted? The CRC-32 value I get for the network file is 3E7C97AE7F79E6D5, and the SHA-256 value is 112BB1C8CDB58B43324F765461DCA727BD468D610D257203E0D950515C15CFCA.

Just refreshed my fork last night - no issues here with Fritz 17 - but I notice some reports on Github:
https://github.com/official-stockfish/S ... ssues/2963

This is NOT an official clone as the bench signature is not aligned, but I don't think you will mind.... ;>)

Code: Select all

I
PGN File: c:/cluster.mfb/pgn/08100236.pgn
Time Control: Time Control-> base+inc: 50+1.000
Games: 20000
Threads: 1
Hash: 512

Current date : time (EDST)
Date: 08/10/20 : 10:16:43

Projected-> Time: 21h:50m:56s
     Run -> Time: 7h:40m:9s

7092 game(s) loaded
Rank Name            Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR
---------------------------------------------------------------------------------------------------------

   1 SF-XIr4-1351     3520   0.0    8    8  2836 1518.5  53.5  758  557 1521  26.7  53.6  3495
   2 Blue-XIr4-1351   3512   8.1    9    9  2838 1478.0  52.1  742  624 1472  26.1  51.9  3497
   3 stockfish-1351   3510   1.7    8    8  2838 1470.5  51.8  716  613 1509  25.2  53.2  3497
   4 Ho-XIr4-1351     3508   2.0    9    9  2836 1459.5  51.5  706  623 1507  24.9  53.1  3498
   5 BD-XIr4-1351     3450  58.3    8    8  2836 1165.5  41.1  436  941 1459  15.4  51.4  3513
---------------------------------------------------------------------------------------------------------

  Δ = delta from the next higher rated opponent
  # = number of games played
  Σ = total score, 1 point for win, 1/2 point for draw

LOS:
                SF Bl st Ho BD
SF-XIr4-1351       92 96 98100
Blue-XIr4-1351   7    62 74100
stockfish-1351   3 37    63100
Ho-XIr4-1351     1 25 36   100
BD-XIr4-1351     0  0  0  0

7092 game(s) loaded

loops/scheduled: 67/200

https://www.dropbox.com/t/lschiRUvc3Vw5WoM

To make this compliant with the license:
Source:
https://github.com/MichaelB7/Stockfish/tree/honey

Thanks for the link to the github discussion, and your fork.

Turns out that on the abrok build I don't even have to rename the file to nn.bin, on Fritz 17 gui I just have to point the evalfile path directly to it like this: ./eval/nn-112bb1c8cdb5.nnue

Also found out that I can just leave it on default path if I put the network file in the same folder as the exe, don't know when this change happened, my apologies if this change was announced I didn't see the announcement.

[gotogo]

[/quote]

For some strange reason the hybrid mode works fine when I rename the network file to nn.bin, and change the evalfile path to ./eval/nn.bin
[/quote]


If you don't mind where do you make these changes?
do you do this before compiling?
Thanks for your time Bob

[mwyoung]

I'm having an issue getting today's update of Stockfish to run in hybrid mode (NNUE enabled). I'm trying to get the one with the timestamp of 1597063097 to work and it's not working for me on either Aquarium or Fritz 17 GUI. I use the build intended for haswell CPU's running on windows, which has the .bmi at the end of the filename. On Fritz 17 GUI the engine loads but when I hit infinite analysis it gives me an error "Stockfish 100820" caused an exception. This is a problem in the engine."

However when I disable hybrid mode the engine runs fine on both GUI's. Is the network file nn-112bb1c8cdb5.nnue corrupted? The CRC-32 value I get for the network file is 3E7C97AE7F79E6D5, and the SHA-256 value is 112BB1C8CDB58B43324F765461DCA727BD468D610D257203E0D950515C15CFCA.

I was having issue to get the NNUE to run without crashing. The good news is the latest release has no issues, and is working fine.

[zullil]

For some strange reason the hybrid mode works fine when I rename the network file to nn.bin, and change the evalfile path to ./eval/nn.bin

If you don't mind where do you make these changes?
do you do this before compiling?
Thanks for your time Bob

EvalFile is a UCI option. It should be visible in any GUI. Put the name of the network file there. And then put the network file in the same directory as the stockfish executable.

No need to alter source code and recompile.

[corres]

...
For some strange reason the hybrid mode works fine when I rename the network file to nn.bin, and change the evalfile path to ./eval/nn.bin

It seems to me the connection between Stockfish-NNUE and NNUE-net is unstable in some measure.
The stability can be enhanced if the name of NNUE-net is standardized (nn.bin) and the connection is hard-coded