Can somebody do simply this? By far the strongest engine!

[corres]

The chimera "engines" are interested by those people who like to make play chess engines instead of themselves and the win is more important for them than the chess game itself.
I think the correct using of two or more chess engine is when the user people act as a supervisor over the analysis of chess engines.

[crem]

Lc0 is ridiculous in endgames, it always was and will probably improve only a little. In 200 games against Fritz 5.32, SF_11 scored 200/200, while Lc0 SV_3200 scored 193/200, Lc0 failing to convert in endgames 14 totally won positions even against Fritz 5.32 !!! It's ridiculous.

Do you still have those games where Lc0 failed to convert? Would be nice to have them for debugging.

[pohl4711]

Thanks, that's it! Works in Cutechess-CLI, now performing a sanity check to see whether I set the options correctly and whether it doesn't have a large overhead.

Please report here about your experiences of Chimera in cutechess-cli. Perhaps, I can do a testrun for my NN vs SF performance-testing in 2 weeks or so, using Chimera with lc0 and Komodo vs. Stockfish and see, if there is an Elo-gain compared to lc0 alone...

https://www.sp-cc.de/nn-vs-sf-testing.htm

It doesn't work as described in Fritz GUI too. Showing weird NPS from move one which tend to converge to lower Lc0 NPS after some thinking, nothing happens at set to switch to SF move 40, same mainly Lc0 all the way (by GPU usage as well).

I dont know, what you did wrong. But here on my machine in Fritz 17 GUI (latest update) Chimera works pretty well. Tried lc0 (0.24.1CUDA)/Stockfish 11 vs. Komodo in an engine-match. For test, it should switch from lc0 to SF after 20 moves. The only small "bug", I could see, was, that SF was starting playing at move 22, not 21. But nodes per second were perfect for lc0 and SF and all worked fine!
As I said, on playchess, some users play with Chimera (I know, a german guy does so, and he is a long time on playchess and takes "his" Elo there very serious and he would never use anything there, which does not work!).

[Laskos]

Lc0 is ridiculous in endgames, it always was and will probably improve only a little. In 200 games against Fritz 5.32, SF_11 scored 200/200, while Lc0 SV_3200 scored 193/200, Lc0 failing to convert in endgames 14 totally won positions even against Fritz 5.32 !!! It's ridiculous.

Do you still have those games where Lc0 failed to convert? Would be nice to have them for debugging.

I have reinstalled windows, those are probably lost, but here are 3 draws against Fritz 6 from a month ago or so. The Lc0 was v024.0 net SV_3200 (large and very strong net), time control 1m + 1s. 2 games show a 50-move draw rule bug, one shows a totally won endgame which Lc0 failed to convert. Hardware is RTX 2070 vs 1 core Fritz 6. No TBs and no any sort of adjudication. I think with Lc0 one should never adjudicate based on anything, it can always fail to convert elementary wins, one never knows.

https://filebin.net/uuahx3g3mtfqno2q

[Laskos]

Thanks, that's it! Works in Cutechess-CLI, now performing a sanity check to see whether I set the options correctly and whether it doesn't have a large overhead.

Please report here about your experiences of Chimera in cutechess-cli. Perhaps, I can do a testrun for my NN vs SF performance-testing in 2 weeks or so, using Chimera with lc0 and Komodo vs. Stockfish and see, if there is an Elo-gain compared to lc0 alone...

https://www.sp-cc.de/nn-vs-sf-testing.htm

It doesn't work as described in Fritz GUI too. Showing weird NPS from move one which tend to converge to lower Lc0 NPS after some thinking, nothing happens at set to switch to SF move 40, same mainly Lc0 all the way (by GPU usage as well).

I dont know, what you did wrong. But here on my machine in Fritz 17 GUI (latest update) Chimera works pretty well. Tried lc0 (0.24.1CUDA)/Stockfish 11 vs. Komodo in an engine-match. For test, it should switch from lc0 to SF after 20 moves. The only small "bug", I could see, was, that SF was starting playing at move 22, not 21. But nodes per second were perfect for lc0 and SF and all worked fine!
As I said, on playchess, some users play with Chimera (I know, a german guy does so, and he is a long time on playchess and takes "his" Elo there very serious and he would never use anything there, which does not work!).

Interesting, I am using a very old Fritz GUI, should try something newer. The same was in Cutechess-CLI at short TC of 15s+0.25s, the playing engine was always the same (the first). Maybe I am doing something systematically wrong.

[Laskos]

Just basic features, no ponder, does not reply when thinking. Supports movetime and wtime btime winc binc. Switching is controlled by game move number.

HeadsUp = Lc0 25.0 blas, up to move 15 and Stockfish 11 for the rest of moves.
TC: Bullet 1'+1"

Result so far after 10 games, using noomen 3 moves opening on cutechess-cli.

Code: Select all

Score of HeadsUp vs Stockfish 11: 2 - 1 - 7  [0.550] 10
...      HeadsUp playing White: 2 - 0 - 3  [0.700] 5
...      HeadsUp playing Black: 0 - 1 - 4  [0.400] 5
...      White vs Black: 3 - 0 - 7  [0.650] 10
Elo difference: 34.9 +/- 121.9, LOS: 71.8 %, DrawRatio: 70.0 %

headsup.cfg

Code: Select all

[engine1]
engine1file = D:\Chess\Engines\Lc0\lc0-v0.25.0-windows-cpu-openblas\lc0.exe
WeightsFile = D:\Chess\Engines\Lc0\lc0-v0.25.0-windows-cpu-openblas\591226.pb.gz
Threads = 2
MinibatchSize = 8
MaxPrefetch = 0

[engine2]
engine2file = D:\Chess\Engines\Stockfish\stockfish_11.exe
Hash = 128

[option]
move_number_switch = 16

Will release tomorrow after some test.

Thanks Ferdy, I was going to write you, you always provided tools for experiments. Could you make an EXE basic UCI compatible wrapper? There are plenty of folks, say in Playchess Engine Room, who need a very strong combo. Take your time to have it work nicely.

Thanks!

[pohl4711]

Older FritzGUIs have problems with lc0. Especially all version before v14.
Because they do not wait long enough, when the engine is loaded, to make sure, that lc0 initializes the NN... And the nodes per second numbers, which are displayed, are bogus, sometimes...
But that happens with lc0 without Chimera, too.

I will try cutechess on my RTX-machine, when the testruns of 3 older nets are finished. In 6-7 days from now. Will report here.
But, as I said, what runs in FritzGUI should run in cutechess-cli, too. I tried 1 minute + 1 second timecontrol by the way.

[Laskos]

Older FritzGUIs have problems with lc0. Especially all version before v14.
Because they do not wait long enough, when the engine is loaded, to make sure, that lc0 initializes the NN... And the nodes per second numbers, which are displayed, are bogus, sometimes...
But that happens with lc0 without Chimera, too.

I will try cutechess on my RTX-machine, when the testruns of 3 older nets are finished. In 6-7 days from now. Will report here.
But, as I said, what runs in FritzGUI should run in cutechess-cli, too. I tried 1 minute + 1 second timecontrol by the way.

Weird, I cannot make it function properly in Fritz 15 GUI. First, here is the XML file created:

Code: Select all

<ChimeraUciConfig SwitchingOverheadSeconds="30" ShowThreads="false" ShowHash="false">
	<Name>Chimera</Name>
	<Engine FromMove="0">
		<Exe>C:\Users\Kai\Downloads\LC0_CUDA_3\lc0_v251.exe</Exe>
		<Arguments />
		<Username />
		<Password />
		<Param Name="WeightsFile">256x20-t40-1541.pb.gz</Param>
	</Engine>
	<Engine FromMove="36">
		<Exe>C:\Users\Kai\Downloads\LC0_CUDA_3\komodo_14.exe</Exe>
		<Arguments />
		<Username />
		<Password />
		<Param Name="Threads">4</Param>
	</Engine>
</ChimeraUciConfig>

Here is what Chimera.exe shows when I run it and then type UCI:

Code: Select all

Eman Chimera 3.0
By: Khalid Omar

[From move:  0] --> Engine: C:\Users\Kai\Downloads\LC0_CUDA_3\lc0_v251.exe
[From move: 36] --> Engine: C:\Users\Kai\Downloads\LC0_CUDA_3\komodo_14.exe

info string Starting engine 1: C:\Users\Kai\Downloads\LC0_CUDA_3\lc0_v251.exe
Loading weights file from: 256x20-t40-1541.pb.gz
Creating backend [cudnn-auto]...
Switching to [cudnn-fp16]...
CUDA Runtime version: 10.0.0
Cudnn version: 7.6.5
WARNING: CUDNN Runtime version mismatch, was compiled with version 7.4.2
Latest version of CUDA supported by the driver: 11.0.0
GPU: GeForce RTX 2070
GPU memory: 8 Gb
GPU clock frequency: 1620 MHz
GPU compute capability: 7.5
uci
id name Chimera
id author Khalid Omar

option name Ponder type check default false
uciok

info string [From move:  0] --> Engine: C:\Users\Kai\Downloads\LC0_CUDA_3\lc0_v251.exe
info string [From move: 36] --> Engine: C:\Users\Kai\Downloads\LC0_CUDA_3\komodo_14.exe

With SF it's identical, so it's not something wrong with Komodo.
Fritz 15 refuses to switch to Komodo from move 36 or whatever to the end.

EDIT: What are these "Username" and "Password"?

[Tibono]

Hi Laskos,
Eman Chimera can fetch each engine from the usual engine's path.
You seem to have configured it to fetch Komodo from LC0's directory, seems a bit strange to me. Did you do it intentionaly, and did you copy the Komodo ".exe" file there? Does Komodo 14 work if you launch it from this LC0 directory?

[Laskos]

Hi Laskos,
Eman Chimera can fetch each engine from the usual engine's path.
You seem to have configured it to fetch Komodo from LC0's directory, seems a bit strange to me. Did you do it intentionaly, and did you copy the Komodo ".exe" file there? Does Komodo 14 work if you launch it from this LC0 directory?

Sure engine is there and works. First I made it from other folders, but I thought that's the problem and moved all onto one sole folder, to no avail.