TalkChess.com

Sylwy wrote: ↑

Tue Dec 01, 2020 8:58 pm

(...)
Next: Orion 0.8 x64 pop-avx-fma Cerebrum by David Carteau (France).

Thanks in advance for testing Orion ! I'm proud of at least one thing : the Cerebrum logo

I see that you removed syzygy support. Did you do anything special with regards to training in late endgame positions? I have played with Stockfish's NNUE nets and code in my own weak engine, and found that it shuffled in some late endgame positions, though that might be due in part to lack of sear...

Sorry for the trouble... That's really nice, thanks a lot ! And don't worry, I'll be silent

Thanks a lot to both of you for your kind words. The training of a NN is a real challenge ! @Gabor : this time, I promise, there won't be a "@CCRL team : oh no, no, no !" topic, as this is an "official" version of Orion :) @Frank : yes I saw that is was too late, but I wanted to wait for the end of ...

After all my NNUE experiments, it's time to release a new version of my little engine Orion ! The main changes are : - the use of a "NNUE-like" neural network as evaluation function ; - the release of both my trainer (Python) and inference (C langage) source code as a "library". More information on ...

Arasan is MIT-licensed: https://en.wikipedia.org/wiki/MIT_License It is a simple but not copyleft license. Thanks Jon for your reply. I was wondering what was the most appropriate license to share my work without implying any restriction on usage by others (I didn't have the NNUE idea, I just imple...

The inference code is quite simple, since you understood what you have to do. It took me only few days (back in August) to have my own C implementation. At the opposite, the training part is highly more difficult to write from scratch... I finally managed to do it a few days ago, with great results ...

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...

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...

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 !