Skip to content

Remove explicit dependence on psqt head in nnuecomplexity - #6631

Open
sscg13 wants to merge 5 commits into
official-stockfish:masterfrom
sscg13:remove-psqt-nnuecomplexity
Open

Remove explicit dependence on psqt head in nnuecomplexity#6631
sscg13 wants to merge 5 commits into
official-stockfish:masterfrom
sscg13:remove-psqt-nnuecomplexity

Conversation

@sscg13

@sscg13 sscg13 commented Feb 23, 2026

Copy link
Copy Markdown
Contributor

Also will make it simpler to test simplifications of psqt head in the future.

Passed non-regression STC:
LLR: 2.93 (-2.94,2.94) <-1.75,0.25>
Total: 97024 W: 25319 L: 25165 D: 46540
Ptnml(0-2): 346, 11411, 24820, 11613, 322
https://tests.stockfishchess.org/tests/view/6997c023742628af3a802bb3

Passed non-regression LTC:
LLR: 2.97 (-2.94,2.94) <-1.75,0.25>
Total: 108384 W: 27844 L: 27711 D: 52829
Ptnml(0-2): 59, 11809, 30326, 11936, 62
https://tests.stockfishchess.org/tests/view/69998392403ae8f29a7fc8bc

bench: 2864296

Co-authored-by: Joost Vandevondele Joost.VandeVondele@gmail.com

@anematode

Copy link
Copy Markdown
Member

Goat

bench: 2864296
@vondele

vondele commented Feb 23, 2026

Copy link
Copy Markdown
Member

one disadvantage of this patch is that the psqt + positional are now combined with prefactors in the network evaluation, so that eval won't directly match the one we get in nnue-pytorch. Making it a bit more difficult to check. I wonder if this term could be directly included in the network training?

@sscg13

sscg13 commented Feb 23, 2026

Copy link
Copy Markdown
Contributor Author

I do think it is odd that using different weightings than in the trainer gains elo, it is something to investigate.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants