All pull requests

Pull requests

• docs: note that random.normal is not bit-exact across GPU architectures

  #4810·electron-rare opened Oct 31, 2026·

  ·

• Fix row offsets and partition stride in GPU sort

  #4807·ArupKantiDas opened Oct 31, 2026Contributor·

  ·
  1

• [CPU] Fix FP4 tie rounding on CPU

  #4805·404khai opened Oct 30, 2026·

  ·
  1

• [CPU] Fix accumulation precision in affine quantized matmul

  #4803·erwinzhang7 opened Oct 30, 2026Contributor·

  ·

• [Metal] Add affine qqmm with int8 NAX MMAs

  #4802·XXXXRT666 opened Oct 30, 2026Contributor·

  ·

• Security hardening for GitHub Actions

  #4801·madrob opened Oct 30, 2026Contributor·

  ·
  1

• [Metal] Route qqmm without global scales like quantized_matmul

  #4796·XXXXRT666 opened Oct 29, 2026Contributor·

  ·

• Check output size for 64-bit indexing in GPU SliceUpdate

  #4794·arian262785-cell opened Oct 29, 2026Contributor·

  ·
  1

• [Metal] Prefer qmm_nax over legacy split-k on NAX devices

  #4793·arian262785-cell opened Oct 29, 2026Contributor·

  ·
  1

• Respect MLX_ENABLE_TF32 in gated_delta_update

  #4792·zcbenz opened Oct 29, 2026Member·

  ·
  1

• Validate trace input ids match trace inputs in import_function

  #4790·turingH opened Oct 29, 2026Contributor·

  ·

• Keep double precision in CPU exp, sin, cos, erf

  #4789·mycomputercrashsalot opened Oct 29, 2026·

  ·

• [Metal] Add affine qmv_mma for 3-16 rows

  #4787·ahuzmeza opened Oct 28, 2026·

  ·

• metal: read FP16 and BF16 affine constants directly for FP32 inputs

  #4785·bri-prism opened Oct 28, 2026Contributor·

  ·
  1

• [CUDA] Sequential kernel for gated_delta_update

  #4783·zcbenz opened Oct 28, 2026Member·

  ·

• Speed up QMM & QMV

  #4781·XXXXRT666 opened Oct 28, 2026Contributor·

  ·
  1

• Bigger MoE tiles on M1 GPUs

  await response
  #4754·a3rosol opened Oct 23, 2026·

  ·
  1

• feat(nn): add freqs and partial_rotary_factor support to RoPE

  low priority
  #4732·shadyuwugurl opened Oct 20, 2026·

  ·

• Add optional int8 path for quantized matmul with affine weights

  await verification
  #4722·magnuslundstedt opened Oct 18, 2026Contributor·

  ·
  1

• Fix custom_function vmap and jvp traversal with opaque bodies

  low priority
  #4612·axiom-of-choice opened Oct 2, 2026Contributor·1 task done·

  ·
  1

  2

• Support uneven tensor parallel sharding in distributed linear layers

  await verificationlow priority
  #4465·twallgren opened Sep 4, 2026·

  ·

• [CUDA] Grouped mm

  await response
  #4390·nastya236 opened Aug 24, 2026Collaborator·

  ·
  1

• Add nanobind 3.0 split-mode support

  await discussion
  #4359·XXXXRT666 opened Aug 19, 2026Contributor·4 tasks done·

  ·
  2

• Add graph-based GPU Timer API

  await discussionlow priority
  #4050·XXXXRT666 opened Aug 7, 2026Contributor·4 tasks done·

  ·

• [CUDA][Improvement] Fast path for gather

  await response
  #3794·nastya236 opened Jul 3, 2026Collaborator·

  ·
  3