Update default net to nn-c38c3d8d3920.nnue

[linrock]

This was a later epoch from the same experiment that led to the previous master net.
After training, it was prepared the same way:

1. greedy permuting L1 weights with Permute master net weights to increase sparsity #4620
2. leb128 compression with Support for LEB128 compression of feature transformer parameters. nnue-pytorch#251
3. greedy 2- and 3- cycle permuting with Update default data to nn-a3d1bfca1672.nnue #4640

Local elo at 25k nodes per move (vs. L1-1536 nn-fdc1d0fe6455.nnue):
nn-epoch739.nnue : 20.2 +/- 1.7

Passed STC:
https://tests.stockfishchess.org/tests/view/64a050b33ee09aa549c4e4c8
LLR: 2.95 (-2.94,2.94) <0.00,2.00>
Total: 195552 W: 49977 L: 49430 D: 96145
Ptnml(0-2): 556, 22775, 50607, 23242, 596

Passed LTC:
https://tests.stockfishchess.org/tests/view/64a127bd3ee09aa549c4f60c
LLR: 2.94 (-2.94,2.94) <0.50,2.50>
Total: 235452 W: 60327 L: 59609 D: 115516
Ptnml(0-2): 119, 25173, 66426, 25887, 121

bench 2427629

[vondele]

hmm for whatever reason the CI is failing again on MacOS.

I wonder if something is non-standard on our regexp that it fails to match bench 2427629 on MacOS ?

[Sopel97]

have you tried skipping the first step and only relying on "greedy 2- and 3- cycle permuting with #4640" ?

official-stockfish/nnue-pytorch#254

[linrock]

have you tried skipping the first step and only relying on "greedy 2- and 3- cycle permuting with #4640" ?

i have, it completely froze up on a few occasions and i don't know why. meanwhile, i haven't run into issues running the 3 steps sequentially other than it being resource intensive and taking about an hour per net.

i can give official-stockfish/nnue-pytorch#254 a try later.

[ppigazzini]

hmm for whatever reason the CI is failing again on MacOS.

I wonder if something is non-standard on our regexp that it fails to match bench 2427629 on MacOS ?

Ubuntu extract the same bench

[vondele]

yeah, we discussing on discord.
It is the order of commits as extracted with the loop that is unexpected

[ppigazzini]

In local test with this branch the script works fine:
bash --noprofile --norc -eo pipefail ../bench.sh
benchref=2427629
From commit: 9f4cd3b
Reference bench: 2427629

[vondele]

it might be related to how this branch is merged in master during the CI ?

[vondele]

look at the additional output in https://github.com/official-stockfish/Stockfish/actions/runs/5478194778/jobs/9978344031
(under Extract bench .... step)