Repository navigation
Releases
Freed V1 (v1.0.2)
A GUI update for Freed V1. (v1.0.1 was skipped: its GitHub release was published before its files were attached and could not be fixed.) The engine is unchanged: the search DLL (32C32302…7E79), endgame DLL (EB8AF045…F0B1) and NNUE model (07B9AB0F…164C) are byte-identical to v1.0.0, so playing strength is the same.
เอนจิ้นหมากฮอสไทย สำหรับคนไทย เพื่อคนไทย โดยคนไทย
What's new
- The first mover is shown as red (แดง เดินก่อน), and the second mover as black.
- Free Threads and Hash entry. The settings now take any whole number, not a fixed list: Threads from 1 to 12 and Hash from 4 to 1024 MB. Values outside that range are clamped, and invalid entries keep the previous value.
Downloads
| File | What it is |
|---|---|
Freed-v1.0.2-windows-x64.zip |
Standalone Windows app (no Python needed) |
freed-1.0.2-py3-none-any.whl |
Python package: pip install it, then run freed --help |
freed-1.0.2.tar.gz |
Python source distribution |
SHA256SUMS.txt |
Checksums for the files above |
The tablebases did not change, so they stay on the v1.0.0 release.
Windows setup
- Extract the zip.
- From the v1.0.0 release, download
tb4_np80.compact.bin,tb5_np80.compact.binand the three TB6 parts. Put them in thebuildfolder insideFreed-v1.0.2. - Join TB6 in that folder:
Then delete the parts. If you already have v1.0.0 set up, you can copy its
copy /b tb6_np80.compact.bin.part1+tb6_np80.compact.bin.part2+tb6_np80.compact.bin.part3 tb6_np80.compact.binbuild\tb*_np80.compact.binfiles instead. - Run
Freed.exe. The GUI opens at http://localhost:8000.
Settings saved by v1.0.0 in the browser carry over.
Freed V1 (v1.0.0)
The first official release of Freed, a rules-exact Thai Makhos (Thai checkers) engine made by Thai people, for Thai people.
เอนจิ้นหมากฮอสไทย สำหรับคนไทย เพื่อคนไทย โดยคนไทย
Highlights
- Native C alpha-beta search with multi-threaded root search
- NNUE evaluation (
model_a.mknnue, SHA-25607B9AB0F…164C) - Exact 3–6 piece endgame tablebases under the 80-ply rule
- Browser GUI with take-back, adjudication and JSON game records
This release ships the gated V1 engine binaries (search DLL 32C32302…7E79, endgame DLL EB8AF045…F0B1).
A rebuild of the search DLL from this tag gave the same move, score and node count as the shipped DLL in all 24 cases of a fixed-depth bench.
Downloads
| File | What it is |
|---|---|
Freed-v1.0.0-windows-x64.zip |
Standalone Windows app (no Python needed) |
freed-1.0.0-py3-none-any.whl |
Python package: pip install it, then run freed --help |
freed-1.0.0.tar.gz |
Python source distribution |
tb4_np80.compact.bin, tb5_np80.compact.bin |
4- and 5-piece tablebases |
tb6_np80.compact.bin.part1 … part3 |
6-piece tablebase in three parts (GitHub's 2 GiB file limit) |
SHA256SUMS.txt |
Checksums for every file |
Windows setup
- Extract the zip.
- Put
tb4_np80.compact.bin,tb5_np80.compact.binand the three TB6 parts into thebuildfolder insideFreed-v1.0.0. - Join TB6 in that folder:
Then delete the parts. The joined file must have SHA-256
copy /b tb6_np80.compact.bin.part1+tb6_np80.compact.bin.part2+tb6_np80.compact.bin.part3 tb6_np80.compact.bin17A0196F3AFEEFD9D57FB00B641045B0EF790B38A64DA73E19DA3F3369D04A4F. - Run
Freed.exe. The GUI opens at http://localhost:8000.
The app refuses to start, with a message, if a tablebase is missing.
Notes
- Tablebase results assume the 80-ply no-progress rule and ignore repetition.
- Windows may warn about an unsigned app on first run.
- Licensed under GPL-3.0.