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
+36-4Lines changed: 36 additions & 4 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -62,9 +62,10 @@ combines established concepts, as the name suggests.
62
62
### Evaluation
63
63
64
64
-[NNUE](https://www.chessprogramming.org/NNUE)
65
-
-`(768 -> 128)x2 -> 1`
66
-
- Quantization to `i16`
67
-
- Trained using original data generated through self-play, starting from a randomly initialized network
65
+
- Architecture: `(768 -> 128)x2 -> 1`
66
+
- Activation Function: `CReLU`
67
+
- Quantization: `i16` (`256`/`64`)
68
+
- Trained on original data generated entirely through self-play
68
69
69
70
## Rating
70
71
@@ -88,6 +89,15 @@ combines established concepts, as the name suggests.
88
89
89
90
You can download precompiled builds from the [GitHub Releases page](https://github.com/codedeliveryservice/Reckless/releases).
90
91
92
+
-`x86_64-v1`: Slowest, compatible with any x86-64 CPU.
93
+
-`x86_64-v2`: Faster, requires support for `POPCNT`, `SEE3`, etc.
94
+
-`x86_64-v3`: Even faster, requires support for `AVX2`, etc.
95
+
-`x86_64-v4`: Fastest, requires support for `AVX512`.
96
+
97
+
For detailed information on the specific features needed for each level, refer to the [x86-64 microarchitecture levels][microarchitecture] Wikipedia page.
0 commit comments