The Leela project has unleashed a graph search improvement at CCC. The result is a bump of maybe +25 uho elo.
Up until recently, most lc0 improvements have been in net architecture. Better net == better data. This also helps Stockfish, which benefits from the better data as well.
The search improvement may not result in massively better data, however, just better search at longer TC. Is this the first lc0 improvement that isn’t matched by an SF improvement? I’ll certainly be curious if and how DAG affects leela training data.
A Leela improvement that won't help Stockfish?
Moderators: hgm, Dann Corbit, Harvey Williamson
-
dkappe
- Posts: 1620
- Joined: Tue Aug 21, 2018 7:52 pm
- Full name: Dietrich Kappe
A Leela improvement that won't help Stockfish?
Fat Titz by Stockfish, the engine with the bodaciously big net. Remember: size matters. If you want to learn more about this engine just google for "Fat Titz".