All pull requests

Pull requests

• Migrate demux to the Backend API

  #2482·mooskagh opened Oct 10, 2026Member·

  ·
  4

• Remove terminal-draw pair branch from best-child comparator

  #2481·mooskagh with CopilotAI opened Oct 10, 2026·

  ·
  1

  

• Convert FEN fullmove numbers to elapsed game plies

  #2480·mooskagh with CopilotAI opened Oct 10, 2026·

  ·
  1

  8

  

• neural: remove dead LoadNetwork and BackendConfiguration

  #2479·mooskagh opened Oct 8, 2026Member·

  ·

• backendbench: do not count the warm-up batch in --batches

  #2478·ContradNamiseb opened Oct 7, 2026Contributor·

  ·
  2

• onnx: copy the position encoding table row by row

  #2477·ContradNamiseb opened Oct 7, 2026Contributor·

  ·

• search: allocate NodeToProcess::eval only for visits

  #2475·ContradNamiseb opened Oct 5, 2026Contributor·

  ·
  2

• Separate the backend into a child process

  demorfc
  #2474·ContradNamiseb opened Oct 2, 2026Contributor·

  ·

• sycl: document the Level Zero v2 adapter workaround on Windows

  #2473·ContradNamiseb opened Oct 2, 2026Contributor·

  ·
  2

• WIP search: AoS picking cache for classic and dag_classic

  wip
  #2468·ContradNamiseb opened Sep 27, 2026Contributor·

  ·

• Backend Pybind11 improvements and interoperability with python-chess

  #2467·saithepaithewhyyy opened Sep 25, 2026Contributor·

  ·
  3

• sycl: fix the linker swap in sycl_build_hack.py for current meson

  #2466·ContradNamiseb opened Sep 19, 2026Contributor·

  ·

• build: oneAPI 2026.1 mkl_sycl_blas_dll lookup fallback (sycl= l0)

  #2465·ContradNamiseb opened Sep 19, 2026Contributor·

  ·
  11

• ci: add GitHub Actions CI for Linux, AMD ROCm/HIP, Windows, and macOS

  #2464·jaipkapoor99 opened Sep 18, 2026Contributor·

  ·
  18

• cublas workaround for fp16 on tu11x is still needed

  #2463·borg323 opened Sep 18, 2026Member·

  ·
  1

• Move onnx caches to user cache directory

  #2461·Menkib64 opened Sep 13, 2026Contributor·

  ·
  1

  11

• xla: add batch bucketing and tpu preset

  #2456·lucario6607 opened Sep 10, 2026·

  ·

• Tpu nan and crash fixes.

  #2453·lucario6607 opened Sep 8, 2026·

  ·
  4

• Geometric policy blending on cuda backend

  #2440·Kovax007 opened Aug 9, 2026Contributor·

  ·
  7

• Add fixed-depth minimax to value-head search

  #2434·natsu0529 opened Jul 31, 2026·

  ·

• classic: Calculate cpuct using started visits

  #2431·Menkib64 opened Jul 21, 2026Contributor·

  ·
  6

• Save TensorRT engine in weights file with onnx2leela and automate during runtime

  #2428·saithepaithewhyyy opened Jul 13, 2026Contributor·

  ·
  94

• Add Support for RMSNorm and optional QKV Biases to Onnx backends

  #2427·john-sp opened Jul 11, 2026Member·

  ·
  5

• Allow multithreaded inference using GraphExecutor

  #2424·almaudoh opened Jun 30, 2026Contributor·

  ·
  11

• Fix rescorer castling and promotion encoding for NNUE data

  #2422·xu-shawn opened Jun 21, 2026·

  ·