Search found 88 matches

by David Carteau
Tue Nov 17, 2020 8:33 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Pytorch NNUE training
Replies: 58
Views: 4666

Re: Pytorch NNUE training

Yes, this comes from using the GPU (hooray pytorch!). As well, Sopel implemented a super fast C++ data parser, which feeds the inputs to pytorch as sparse tensors, which was a very large speedup (since the inputs to the first layer are super sparse). What was the batch size for this? That makes a b...
by David Carteau
Tue Nov 17, 2020 8:24 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Pytorch NNUE training
Replies: 58
Views: 4666

Re: Pytorch NNUE training

For now, I have no gpu, so I run on cpu only. batch_size is 256, 8 workers => it takes more than 1h for 10Msfens Unless "step" in the tensorboard view is not equivalent to sfens ... So question is, do I need a gpu or is there something else going wrong ? Hi Vivien, Until now, I was using only CPU t...
by David Carteau
Sun Nov 15, 2020 9:07 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Pytorch NNUE training
Replies: 58
Views: 4666

Re: Pytorch NNUE training

gladius wrote:
Sat Nov 14, 2020 9:30 pm
An initial training run by vondele on 256M positions is looking much more promising (also takes about 10 minutes to do an epoch - a pass through the 256M positions, which is great).
Wow, it takes my trainer about... 3 days to perform only one epoch (360M positions) ! Great job !
by David Carteau
Sat Nov 14, 2020 10:13 am
Forum: Computer Chess Club: Tournaments and Matches
Topic: Stargazer under the Rise of Orion (various versions of recently progressing engines)
Replies: 10
Views: 785

Re: Stargazer under the Rise of Orion (various versions of recently progressing engines)

Angle wrote:
Thu Nov 12, 2020 1:57 pm
(...)
Very impressive results shown by Orion 0.7 NNUE (expected).
(...)
01. Orion 0.7 NNUE ---------- 52,5
(...)
33. Orion 0.7 ----------------- 13,0
Thank you for organizing this tournament, and thank you for your interest in my little engine Orion :)
by David Carteau
Sat Nov 14, 2020 10:10 am
Forum: Computer Chess Club: General Topics
Topic: Orion 0.7 : NNUE experiment
Replies: 29
Views: 8984

Re: Orion 0.7 : NNUE experiment

Yeah ! Yeah ! and... Yeah !! As reported in another post , I was struggling last weeks with my neural network trainer. I did not understand why things went so wrong. I first tried a "manual" implementation of weights tuning, with genetic algorithms and then stochastic gradient descent (much more com...
by David Carteau
Wed Nov 04, 2020 5:23 am
Forum: Computer Chess Club: General Topics
Topic: Minic version 3
Replies: 20
Views: 1857

Re: Minic version 3

Second year anniversary for Minic :D ! and a new version number 3.00 Porting Stockfish NNUE implementation inside Minic was fun and instructive. This has clearly raised my interest in NN technologies and in this sense is a success. As you already know, I was never comfortable with the big copy/past...
by David Carteau
Sun Sep 27, 2020 2:01 pm
Forum: Computer Chess Club: Tournaments and Matches
Topic: @CCRL team : oh no, no, no !
Replies: 6
Views: 1587

Re: @CCRL team : oh no, no, no !

(...) As it is now it's impossible to know the origin of the net and to whom goes the credit, except Igel that has its own created net. nn-82215d0fd0df nn-97f742aaefcd If someone knows the right SV net version number I can change it. Orion 0.7 NNUE uses the 82215d0fd0df network provided by Sergio V...
by David Carteau
Sun Sep 27, 2020 7:57 am
Forum: Computer Chess Club: Tournaments and Matches
Topic: @CCRL team : oh no, no, no !
Replies: 6
Views: 1587

Re: @CCRL team : oh no, no, no !

Dear David, I did the test to satisfy my own curiousity. I know it is not official but our list is not exclusive to official versions only, in fact it includes several engines that have never been released. And, on your web page you write: 'I managed to get my own NNUE implementation'. So in fact t...
by David Carteau
Sun Sep 27, 2020 6:46 am
Forum: Computer Chess Club: Tournaments and Matches
Topic: @CCRL team : oh no, no, no !
Replies: 6
Views: 1587

@CCRL team : oh no, no, no !

I just saw that CCRL blitz list had been updated and that "Orion 0.7 NNUE 64-bit" was listed on 41st rank with a rating of 3178, competing with Senpai 2.0. Please, no ! This version is not the "official" Orion, it's only the result of an experiment which had the only aim to make me understand how ne...
by David Carteau
Sat Sep 26, 2020 12:36 pm
Forum: Computer Chess Club: General Topics
Topic: Minic version 2
Replies: 157
Views: 42147

Re: Minic version 2

(...) Well "Napping Nexus" is an experimental first not-too-bad network I train from SF data, using Nodchip SF leaner. It knows how to play, kills Minic standard eval easily but it still very very far from best SV nets ! I'm glad it works, this is a first step in building nets by myself but I expec...