All pull requests

Pull requests

• cmake: avoid linking MATH_LIBRARY-NOTFOUND

  #1647·hoult14 opened Sep 29, 2026·

  ·

• mnist : make mnist_model move-only

  #1640·165hh opened Sep 24, 2026·

  ·
  1

• add getexecpath for get_executable_path() on OpenBSD

  #1631·rfht opened Sep 18, 2026·

  ·

• fix(cuda): include <cuda/execution> and <cuda/stream_ref> for CUB top-k path (CCCL 3.2+)

  #1625·ousamabenyounes opened Sep 11, 2026·

  ·
  1

• simple : temporary change to trigger Vulkan crash on macOS

  #1623·ggerganov opened Sep 9, 2026Member·

  ·
  1

• Fix builds on s390x

  #1620·WhyNotHugo opened Sep 6, 2026·

  ·
  1

• vulkan: only request 32-wide FA subgroups where size control can grant them

  #1619·thewh1teagle opened Sep 4, 2026Contributor·

  ·

• vulkan: fall back to vkGetDeviceQueue when vkGetDeviceQueue2 returns null

  #1616·thewh1teagle opened Sep 4, 2026Contributor·

  ·
  1

• vulkan: load the Vulkan loader at runtime instead of linking it

  #1614·thewh1teagle opened Sep 4, 2026Contributor·

  ·

• ggml-opencl: guard F16 KQ/KQV path with use_adreno_kernels (fix SD3.5 NaN)

  #1613·Hmission opened Sep 2, 2026·

  ·

• ggml-opencl: Mali tiled GEMM half-precision variant (2.86x on Mali-G925)

  #1612·Hmission opened Sep 2, 2026·

  ·

• sgemm: handle m % 4 != 0 instead of declining the whole matmul

  #1604·femelo opened Aug 21, 2026·

  ·

• sgemm: write the operand addresses so GCC will strength-reduce them on aarch64

  #1603·femelo opened Aug 21, 2026·

  ·

• sgemm: don't pick a register tile GCC will spill on aarch64

  #1602·femelo opened Aug 21, 2026·

  ·

• Fix Qualcomm HTP and Vulkan execution on Windows ARM64

  #1601·roozbehid opened Aug 21, 2026·

  ·

• Add WIN_PART/WIN_UNPART + head-size-32 attention for SAM3

  #1590·greenjava opened Aug 18, 2026·

  ·

• fix(metal): do not latch has_error on transient GPU failures

  #1585·olxqnz5et7g2rv opened Aug 16, 2026·

  ·
  1

• Implement GGML_OP_IM2COL_3D for Metal backend

  #1574·venomega opened Jul 27, 2026·

  ·

• Gate RVV code on the target extension

  #1571·catap opened Jul 19, 2026·

  ·
  1

• opencl: add Adreno xmem SDPA path

  #1562·happyyzy opened Jul 11, 2026Contributor·

  ·

• opencl: fix Q4_0 SOA rows and Adreno paths

  #1561·happyyzy opened Jul 11, 2026Contributor·

  ·
  3

• sycl: fix soft_max_f32 max reduction

  #1534·someoneinjd opened Jun 10, 2026Contributor·

  ·
  2

• metal : graceful failure instead of GGML_ABORT

  #1527·fiorelorenzo opened Jun 3, 2026·

  ·

• tests : add OLMoE-sized K-quant shapes to test_mul_mat_id (ref #1506)

  #1518·devYRPauli opened May 28, 2026Contributor·

  ·
  3

• Fix build and test running for Windows MSVC

  #1512·MaxwellGengYF opened May 26, 2026·

  ·