You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
It might grow to be important now that SF16 has normalized the scores. That could mean the progress in leaf evaluation discerning power, and therefore that of the score of the root of the search (attributed from min-maxing the searched tree set of leaf evaluations), would happen in the decimals, if SF development experiments allow to higher resolution at input and output of that function basis (including NNue of course).
Good find.
PS: Anyone knows of the true internal resolution or precision of SF16 (I know wrong repo....) Asking the op, in passing.
Evals in PGN stored with only 1 digit after decimal point, for example:
1.g3 Nc6 {End of openingA00: Benko's opening} 2.d4 {+0.3/21 2424 1511177} d5 {+0.3/24 6528 2751448}
Expected 2 digits like in most other GUIs.
Latest version of GUI - 0.58 RC1
The text was updated successfully, but these errors were encountered: