Nemorino 6 (NNUE)

[dkappe]

Is Nemorino misconfigured at CCCC? A threading issue? It’s not doing too well.

[Graham Banks]

Is Nemorino misconfigured at CCCC? A threading issue? It’s not doing too well.

Did they tick the 'use NNUE' box in the parameters?

[RubiChess]

They are searching for a new TD via bot message so they probably have bigger problems than Nemorino settings.

Andreas

[Madeleine Birchfield]

I just released version 6 of Nemorino!
Main new feature is the support for NNUE networks, which improves Nemorino's playing strength a lot!

Nemorino comes with it's own network. The network format is slightly different than the original format from Stockfish as I removed the 64 "BONA_PIECE_ZERO" input parameters from the weights file and from the inference code (which is an adapted copy of the code from Stockfish). Nevertheless Nemorino can use the network files from Stockfish as well - it will simply create a converted copy on first use.

Additionally as it's more than 2 years since the last release, this version contains a lot of minor improvements and bugfixes.

You can find source code and binaries for windows at https://bitbucket.org/christian_g_nther/nemorino

What is the difference between the Nemorino 6.00 that is publicly available on the linked website and on the ratings list and the Nemorino 6.03 that was playing in Chess.com's Computer Chess Championship and that is playing in TCEC Season 20?

[Florentino]

Nemorino 6:
- Uses a net (nemorino_net_6.nnue) trained with training data generated with Nemorino partly using nets from Stockfish

Nemorino 6.03:
- Uses a net (nemorino_net_13.nnue) trained with training data generated exclusively from Nemorino classic eval
- Uses flip instead of rotate
- Is much weaker than Nemorino 6.00 (but much stronger than any Nemorino 5.* version)
- Was created in relatively short time for TCEC FRC event

[Madeleine Birchfield]

Nemorino 6:
- Uses a net (nemorino_net_6.nnue) trained with training data generated with Nemorino partly using nets from Stockfish

Nemorino 6.03:
- Uses a net (nemorino_net_13.nnue) trained with training data generated exclusively from Nemorino classic eval
- Uses flip instead of rotate
- Is much weaker than Nemorino 6.00 (but much stronger than any Nemorino 5.* version)
- Was created in relatively short time for TCEC FRC event

Something else I would suggest for future Nemorino versions is a rewrite of the neural network inference code from the ground up so that the inference code isn't a copy of Stockfish's code. There are quite a few people around who are unhappy with your current inference code being copied from Stockfish.

[matejst]

Something else I would suggest for future Nemorino versions is a rewrite of the neural network inference code from the ground up so that the inference code isn't a copy of Stockfish's code. There are quite a few people around who are unhappy with your current inference code being copied from Stockfish.

Why should Christian Günther care about making a few (un)happy? The point with open source and GPL is the reuse of code, and the availability of the new code. Nemorino is open source, and the author gave credits where credits were due.

[Chessqueen]

I just released version 6 of Nemorino!
Main new feature is the support for NNUE networks, which improves Nemorino's playing strength a lot!

Nemorino comes with it's own network. The network format is slightly different than the original format from Stockfish as I removed the 64 "BONA_PIECE_ZERO" input parameters from the weights file and from the inference code (which is an adapted copy of the code from Stockfish). Nevertheless Nemorino can use the network files from Stockfish as well - it will simply create a converted copy on first use.

Additionally as it's more than 2 years since the last release, this version contains a lot of minor improvements and bugfixes.

You can find source code and binaries for windows at https://bitbucket.org/christian_g_nther/nemorino

Ethereal should be extremely strong with NNUE

[Damir]

It seems to me that in majority of the posts regarding chess engines Madeleine always jumps to the conclusions and literally talks out of her ass...

[Madeleine Birchfield]

It seems to me that in majority of the posts regarding chess engines Madeleine always jumps to the conclusions and literally talks out of her ass...

This is not true, and this is especially not true in the case of the posts I made on this thread.

1. People above in this thread, such as Dietrich Kappe, have noted that Nemorino 6.03 seems weaker than Nemorino 6.00, and I wanted an explanation why that might be the case.

2. Christian specifically said in Nemorino's Readme.md and in the network.h files that most of his inference code was copied from Stockfish, and the likes of Andrew Grant have been complaining about for the past few weeks here and in the TCEC and OpenBench communities about engines copying Stockfish code and using Stockfish to help train their nets. I wanted to alert Christian to that fact that there are people who disagree with his choices to copy Stockfish code and give him a solution to placate them.