BT5-1024x15x32h-rpe-swa-2647500.pb.gz

[lucario6607]

Only real difference with rpe is the speed and the slightly better policy, otherwise I dont think it does much. There are also quantized bt4 but don't think the speed up was that great due to only a small part of the net being quantized

[peter]

What a great way to discredit yourself, by mentioning shashchess. Funny how its test show it is better than stockfish. You can also simply read the code of the engine

"If you still care, point him to <https://github.com/LeelaChessZero/lc0/b ... /search.cc> where all references to multipv are in the uci info output function."

I mentioned ShashChess just because on its github- site the 256 positions of this one "my" often used suite (I wrote times and times again here and in other fora, how and out of which sources I collected it, in most of the positions the original sources, as far as I know them, are given e.g. in single .epd- string of each one position too) can be downloaded, just as one of the sources, to get it as a complete suite:

https://github.com/amchess/ShashChess/t ... s/2023-256

because Andrea Manzo uses it also, and you see my name mentioned on the site just because I recommended this one suite to him way back then and did some testing once in a while for him too.
Here

viewtopic.php?p=980708#p980708

I gave it latest in this one forum in .epd- format.

I don't claim the suite to be anything very special except it's a minimum of number of positions to be run automatically with certain hardware- TC fitting to a certain pool of engines as for a common level of "difficulty" (hardware- time necessary to get solved for correct reasons with a mimimum of happenstance), mimimum of number of positions and hardware- time as for statistical relevance of the results and error bars of single runs, especially if evaluated as a bigger number of runs engine (-setting) to engine and position to position as well as for solved- unsolved as for time indices with EloStatTS by Frank Schubert.

Again and again I also said it here and in other fora not to be compared as for the results to quite different other one suites (have really a bunch of those as well as single best move- ones as such with multiple solutions and to be used with tools like MEA and NICE, starting form 128 positions up to 2048, each one for certain tools, engines and hardware- TC) especially not with results out of game playing (which depend on pool of engines, hardware- TC and especially test positions of opening as well).

If you think, all this experience of mine with positional testing over the many years I do all this now (might be well over 30 years already, even if I only count the years, I then later on had started with automatically run suites and not all the years before and later on, I did and do evaluate engines, chess- positions, - moves, -variants and -games by their analysis- output at all ), so if you think all these things "discredit me", as you write, it's ok for me. On the other hand, what discredits your opinions about such things to me mainly, is that you are not willing or able to have a look at output- lines of engines at all obviously to get information more than just the move- choice only.

So if you want to stick to eng-eng-game playing only to evaluate their performance chess- wise, it's fine for me, but why then all this discussion with me about things, you aren't interested in at all?
BTW, if you might not have already seen it, I had one more posting to the question of time to solution in MultiPV- mode and with single primary variant to the example of the Salai- position, I had already sent it before your one but latest answer to me, so yet read it still or leave it for good, I for my part and for the discussion with you, I'm outa here now, regards

[Paloma]

Code: Select all

(I wrote times and times again here and in other fora, how and out of which sources I collected it, in most of the positions the original sources, as far as I know them, are given e.g. in single .epd- string of each one position too) can be downloaded, just as one of the sources, to get it as a complete suite:

https://github.com/amchess/ShashChess/t ... s/2023-256

How can I download this suite 2023-256 completely? I only see single files: .cbh, .cba, .cbh, .cbg etc.

[peter]

How can I download this suite 2023-256 completely? I only see single files: .cbh, .cba, .cbh, .cbg etc.

You have to download all of them to have a complete cb- database then in the same directory collected for a for these (Fritz and cb- database)- GUIs readable database.

The much easier way is to download the .epd- files from the other one link in same posting you just answered to. In the older posting from this link you have all the .epd- strings in a code- window. There
CODE: SELECT ALL
- copy and paste into a new Editor- window, store as .epd- instead of .txt- file. In this one (.epd)- format you can easily get a .cbh- file out of it again by clicking the "Paste"- Icon in database- window in Fritz or cb, then submenu "Append EPD" and choose the file you created newly with name you gave it in .epd- format.

[lucario6607]

Code: Select all

(I wrote times and times again here and in other fora, how and out of which sources I collected it, in most of the positions the original sources, as far as I know them, are given e.g. in single .epd- string of each one position too) can be downloaded, just as one of the sources, to get it as a complete suite:

https://github.com/amchess/ShashChess/t ... s/2023-256

How can I download this suite 2023-256 completely? I only see single files: .cbh, .cba, .cbh, .cbg etc.

Why would you want anything from shashchess?

[cc2150dx]

Change the backend to onnx-trt e.g lc0 -b onnx-trt

Tried this with the newly downloaded binary and UCI- command to change backend- type:

Code: Select all

setoption name Backend value onnx-trt
go depth 9
Found pb network file: C:\eng\lc0-v0.32.1-windows-onnx-trt/BT5-1024x15x32h-rpe-swa-3700000.pb.gz
Weights file has multihead format, updating format flag
Creating backend [onnx-trt]...
...
info depth 9 seldepth 32 time 158224 nodes 57102 score cp 11 nps 1998 tbhits 0 pv d2d4 d7d5 c2c4 c7c6 c4d5 c6d5 c1f4 g8f6 e2e3 b8c6 f1b5 c8g4 g1f3 e7e6 d1a4 g4f3
bestmove d2d4 ponder d7d5

So far so well, with UCI option Backend onnx-trt it runs in Fritz 20 too, thanks again. And yes, I know the BT5 net meant for high end GPU- installations only, so I won't give it much trials at the PC it's already running on now, because this one has single Nvidia 3070ti only. When I'll also succeed to install tensorRt at WLAN- connected newer one too, there's a RTX 5070 installed, but I guess, this will be much too slow still, won't it?

Peter, Newer binary here: https://ci.appveyor.com/api/buildjobs/5 ... c0-trt.exe Just rename file to Lc0 for Classic Search or Lc0-dag-preview for dag-preview search. You'll also need newer onnx-runtime dlls https://github.com/microsoft/onnxruntim ... 1.23.2.zip Just swap out your old dllls (4 in total) for the newer ones.

[peter]

Newer binary here: https://ci.appveyor.com/api/buildjobs/5 ... c0-trt.exe Just rename file to Lc0 for Classic Search or Lc0-dag-preview for dag-preview search. You'll also need newer onnx-runtime dlls https://github.com/microsoft/onnxruntim ... 1.23.2.zip Just swap out your old dllls (4 in total) for the newer ones.

Get this one message after starting binary in console:

WARNING: Simplified version without CUDA enhancements.

How to deal with the subdirectories of the --1.23.2.zip- package, extract content to directory of binary too or keep hierarchy as in download?

[cc2150dx]

Don't worry about the warning. Just pick out the 4 dll's from the Lib folder and replace.

[peter]

Ok, just should have waited a little longer probably before engine then started to run anyhow.
Thanks for new version, will give it a try, regards

[cc2150dx]

You're welcome