Repository navigation
Using Stockfish-labelled positions to train a new engine's NNUE: any guidance? #7244
cankasaplar
started this conversation in
General
Replies: 1 comment 1 reply
|
You can label positions by stockfish and train on them, not much particular to do, nice if you acknowledge. Leela licenses its data you can checkout their license. Having a clean pipeline for training and testing is very important to improve in the long run. Fishtest like testing works well, and we've started to declare our training process as .yaml recipe so it can be made reproducible, including the data download steps. |
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Hello everyone,
I'm an independent developer building a UCI engine in Rust (Rhizoh, NNUE-based, playing live at rhizoh.com). Thank you for Stockfish, fishtest and nnue-pytorch, they are the reference we learn from.
We plan to improve our network by generating positions and labelling them with Stockfish, run locally as an external engine. We do not copy Stockfish source code or networks into our engine and we do not distribute Stockfish, so we believe we stay within the GPLv3 terms in the README. Before we start at scale, I'd like to ask:
We will credit Stockfish and the Leela project clearly on our site and are glad to share our measurements back.
Thanks for your time.
Can Kasaplar (rhizoh.com)
All reactions