Update default Network to [DB0BF8E7]

AndyGrant · AndyGrant · commit 14abc7059f33 · 2024-04-21T22:35:09.000-04:00
An additional 71 epochs (1024x1024x128) steps, with 100% WDL, with the addition of S60/S61 data. This data was generated with normalization on, by accident, and was discarded for previous runs as a result. However, with 100% WDL, the only issue is that the openings randomly generated were more extreme than typical, which is fine. It is still unclear why I am unable to train a new network from scratch, which comes even close to master. The leading theory right now is dead neurons. To the Stockfish developers, who read all of my commits, even though I seldom read theirs when it comes to Torch: This elo comes from the inclusion of new data. There is nothing particularly special about the new data. The gains continue with the 100% WDL as seen prior. Elo | 3.54 +- 3.06 (95%) SPRT | 10.0+0.10s Threads=1 Hash=8MB LLR | 2.95 (-2.94, 2.94) [0.00, 3.00] Games | N: 23044 W: 5503 L: 5268 D: 12273 Penta | [123, 2641, 5776, 2842, 140] http://chess.grantnet.us/test/36402/ Elo | 3.91 +- 3.37 (95%) SPRT | 60.0+0.60s Threads=1 Hash=64MB LLR | 2.95 (-2.94, 2.94) [0.00, 3.00] Games | N: 18494 W: 4315 L: 4107 D: 10072 Penta | [17, 2064, 4889, 2248, 29] http://chess.grantnet.us/test/36403/ BENCH 3,171,015
diff --git a/src/uci.h b/src/uci.h
@@ -23,7 +23,7 @@
 
 #include "types.h"
 
-#define VERSION_ID "14.34"
+#define VERSION_ID "14.35"
 
 #ifndef LICENSE_OWNER
     #define LICENSE_OWNER "Unlicensed"
