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
Copy file name to clipboardExpand all lines: README.md
+7-7Lines changed: 7 additions & 7 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -14,7 +14,7 @@ Neural Network Evaluation
14
14
Starting with version 10 release, Tucano uses a neural network evaluation, which increases the engine strength. This network architecture is based on stockfish neural network.
15
15
From version 10.06 and later, including official release 11.00, the network has been changed and use a new structure of 768x512x1 neurons. Also includes its own network trainer. The weights are embeeded in the file eval_nn.h.
16
16
17
-
Starting with version 11.01, it returns to version 10 archictecture, using NNUE evaluation, with incremental update of the position. For this version it will require the network file that can be download from https://github.com/alcides-schulz/TucanoNets.
17
+
Starting with version 11.01, it returns to version 10 archictecture, using NNUE evaluation, with incremental update of the position. For this version it will require the network file that can be download from https://github.com/alcides-schulz/TucanoNets. This file should be put in the same folder as the executable so tucano can find it automatically. Otherwise inform in the parameters as explained below.
18
18
19
19
Terms of use
20
20
------------
@@ -68,6 +68,7 @@ If you compile tucano you can use the command "bench" to get a signature. Just s
68
68
Signature is a number generated after searching a couple of positions to indicate you have the correct compilation.
69
69
If you don't get the correct signature it means that something is wrong with the compilation process and the program may not perform correctly.
70
70
71
+
12.00: 5620048
71
72
11.00: 5203395
72
73
10.00: 5734637 (with nn eval file loaded)
73
74
9.00: 21898211
@@ -100,24 +101,23 @@ You can report issues with other platforms and I will try to address as possible
Note: It is recommend to use AVX2 or SSE4 in order to have a good performance with neural network evaluation. The OLD version is basic and doesn't have the performance benefits of avx2 and sse4 architectures but can work for old plataforms.
121
+
Note: It is recommended to use AVX2 or SSE4 in order to have a good performance with neural network evaluation. The OLD version is basic and doesn't have the performance benefits of avx2 and sse4 architectures but can work for old plataforms.
0 commit comments