9fcd47a717
This network is further trained on a new BT4 distillation stage, fine tuning on ~2 billion positions relabeled with the value head output of `BT4-tf13tune.pb.gz`. The dataset can be found at https://huggingface.co/datasets/xushawn/test80-bt4-relabel. A modified branch of lc0 was used to derive this data: https://github.com/xu-shawn/lc0/tree/relabel_dual_stream_test 2 billion positions represent a tiny subset of the total training data, and BT4 relabeling is inherently computationally expensive. I expect a lot more gains as more data are relabeled, but it will likely require coordinated community effort. Everyone is welcome to contribute, and yl25946 has made a spreadsheet to track progress: https://docs.google.com/spreadsheets/d/1yanofhusEzDg8ZnurAw799ikoTY6GcqsNMYfpswOIbw/edit. Special thanks to Viren6, who performed policy/value distillation experiments on Monty, and created the lc0 distillation fork that the current relabeler is based on; yl25946 for proposing the idea of large network distillations back in February 2025, running distillation experiments on the HL4096 network, and working on fine tuning attempts; vondele for nettest and suggesting the fine-tuning approach; and many others on the knowledge distillation thread in the SF Discord #ideas channel. nettest PR: https://github.com/vondele/nettest/pull/369 Ongoing STC: LLR: -0.01 (-2.94,2.94) <0.00,2.00> Total: 72224 W: 18891 L: 18784 D: 34549 Ptnml(0-2): 336, 8437, 18332, 8798, 209 https://tests.stockfishchess.org/tests/view/6a3ae7913036e45021aeb4a0 Passed LTC: LLR: 2.94 (-2.94,2.94) <0.00,2.00> Total: 25110 W: 6566 L: 6288 D: 12256 Ptnml(0-2): 27, 2625, 6957, 2935, 11 https://tests.stockfishchess.org/tests/view/6a3b73513036e45021aeb51e Passed VLTC: LLR: 2.94 (-2.94,2.94) <0.50,2.50> Total: 18544 W: 4924 L: 4658 D: 8962 Ptnml(0-2): 5, 1730, 5533, 2002, 2 https://tests.stockfishchess.org/tests/view/6a3bbe233036e45021aeb56e closes https://github.com/official-stockfish/Stockfish/pull/6924 Bench: 2710209 Co-authored-by: Li Ying <121075683+yl25946@users.noreply.github.com> Co-authored-by: Viren6 <94880762+Viren6@users.noreply.github.com>
Stockfish
A free and strong UCI chess engine.
Explore Stockfish docs »
Report bug
·
Open a discussion
·
Discord
·
Blog
Overview
Stockfish is a free and strong UCI chess engine derived from Glaurung 2.1 that analyzes chess positions and computes the optimal moves.
Stockfish does not include a graphical user interface (GUI) that is required to display a chessboard and to make it easy to input moves. These GUIs are developed independently from Stockfish and are available online. Read the documentation for your GUI of choice for information about how to use Stockfish with it.
See also the Stockfish documentation for further usage help.
Files
This distribution of Stockfish consists of the following files:
-
README.md, the file you are currently reading.
-
Copying.txt, a text file containing the GNU General Public License version 3.
-
AUTHORS, a text file with the list of authors for the project.
-
src, a subdirectory containing the full source code, including a Makefile that can be used to compile Stockfish on Unix-like systems.
-
a file with the .nnue extension, storing the neural network for the NNUE evaluation. Binary distributions will have this file embedded.
Contributing
See Contributing Guide.
Donating hardware
Improving Stockfish requires a massive amount of testing. You can donate your hardware resources by installing the Fishtest Worker and viewing the current tests on Fishtest.
Improving the code
In the chessprogramming wiki, many techniques used in Stockfish are explained with a lot of background information. The section on Stockfish describes many features and techniques used by Stockfish. However, it is generic rather than focused on Stockfish's precise implementation.
The engine testing is done on Fishtest. If you want to help improve Stockfish, please read this guideline first, where the basics of Stockfish development are explained.
Discussions about Stockfish take place these days mainly in the Stockfish Discord server. This is also the best place to ask questions about the codebase and how to improve it.
Compiling Stockfish
Stockfish has support for 32 or 64-bit CPUs, certain hardware instructions, big-endian machines such as Power PC, and other platforms.
On Unix-like systems, it should be easy to compile Stockfish directly from the
source code with the included Makefile in the folder src. In general, it is
recommended to run make help to see a list of make targets with corresponding
descriptions. An example suitable for most Intel and AMD chips:
cd src
make -j profile-build
Detailed compilation instructions for all platforms can be found in our documentation. Our wiki also has information about the UCI commands supported by Stockfish.
Terms of use
Stockfish is free and distributed under the GNU General Public License version 3 (GPL v3). Essentially, this means you are free to do almost exactly what you want with the program, including distributing it among your friends, making it available for download from your website, selling it (either by itself or as part of some bigger software package), or using it as the starting point for a software project of your own.
The only real limitation is that whenever you distribute Stockfish in some way, you MUST always include the license and the full source code (or a pointer to where the source code can be found) to generate the exact binary you are distributing. If you make any changes to the source code, these changes must also be made available under GPL v3.
Acknowledgements
Stockfish uses neural networks trained on data provided by the Leela Chess Zero project, which is made available under the Open Database License (ODbL).