Tomasz Sobczyk

From Chessprogramming wiki
Jump to: navigation, search

Home * People * Tomasz Sobczyk

Tomasz Sobczyk, (Sopel97, Sopel)
a Polish computer scientist involved in recent Stockfish development and documentation concerning NNUE [1]. He introduced new net architectures using the HalfKA, HalfKAv2 [2] [3] and HalfKAv2_hm [4] [5] feature sets, optimized the NNUE inference code for various SIMD instruction sets [6] [7] [8], and contributed to Gary Linscott's Pytorch NNUE project to train Stockfish, mostly working on optimizations which allowed nets to be trained within hours instead of days [9] [10] [11]. He also heavily contributed to the original NNUE trainer and data generator by Yu Nasu, cleaning up the codebase, updating the data generator, optimizing the trainer, and adding other useful tools [12] which are now available in the tools branch of the official-stockfish repository [13]. He also introduced the Binpack storage format for training data [14] [15], comprising of position evaluations from chess games, which utilizes efficient delta encoding, and reduces the sizes of the datasets by 10 to 20 times compared to the previous solutions. His contributions span more than 200 pull requests over multiple repositories [16] [17] [18]. He's also an author of a chess engine Fat Titz, which is based on CFish and is a parody of Fat Fritz [19] [20].

Forum Posts

External Links

Stockfish

Pytorch NNUE

References

  1. NNUE Guide (nnue-pytorch/nnue.md at master · glinscott/nnue-pytorch · GitHub) hosted by Gary Linscott
  2. Update default net to nn-8a08400ed089.nnue by Sopel97 · Pull Request #3474 · official-stockfish/Stockfish · GitHub
  3. nnue-pytorch/nnue.md at master · glinscott/nnue-pytorch · GitHub - HalfKAv2 feature set
  4. New NNUE architecture and net by Sopel97 · Pull Request #3646 · official-stockfish/Stockfish · GitHub
  5. nnue-pytorch/nnue.md at master · glinscott/nnue-pytorch · GitHub - HalfKAv2 feature set
  6. Optimization of the affine transformations. by Sopel97 · Pull Request #3203 · official-stockfish/Stockfish · GitHub
  7. AVX-512 optimizations. by Sopel97 · Pull Request #3218 · official-stockfish/Stockfish · GitHub
  8. Optimize and tidy up affine transform code. by Sopel97 · Pull Request #3663 · official-stockfish/Stockfish · GitHub
  9. Defer data preparation to native code. Use sparse input tensors. · Pull Request #1 · glinscott/nnue-pytorch · GitHub
  10. A custom kernel for the feature transformer. · Pull Request #96 · glinscott/nnue-pytorch · GitHub
  11. Re: will Tcec allow Stockfish with a Leela net to play? by Connor McMonigle, CCC, June 17, 2021
  12. Pull Requests by Sopel97 · nodchip/Stockfish · GitHub
  13. official-stockfish tools branch · official-stockfish/Stockfish · GitHub
  14. Sopel97/nnue_data_compress · GitHub
  15. Introduce sfen_format option for gensfen. Experimental support for binpack format in gensfen and learn. · Pull Request #129 · nodchip/Stockfish · GitHub
  16. Pull Requests by Sopel97 · nodchip/Stockfish · GitHub
  17. Pull Requests by Sopel97 · official-stockfish/Stockfish · GitHub
  18. Pull Requests by Sopel97 · glinscott/nnue-pytorch · GitHub
  19. Sopel97/FatTitz · GitHub
  20. Fat Titz 1.0 released by Sopel, CCC, Aug 26, 2021

Up one level