Commits

Branch selector

main

User selector

All users

All time

Commit history

Commits on Jul 8, 2026

• Patch bump to 0.32.1 (#3816)

  zcbenz

  authored
  57c66caView commit details

  Copy full SHA for 57c66ca

  Browse repository at this point

Commits on Jul 7, 2026

• Fix conv2 gradients in grouped strided case on Metal (#3800)

  ericphanson

  authored
  7a1d4f5View commit details

  Copy full SHA for 7a1d4f5

  Browse repository at this point

• Fix infinite norm negative axis mismatch bug for matrices with 2 or more dimensions. (#3756)

  danlee2002

  authored
  3a73f21View commit details

  Copy full SHA for 3a73f21

  Browse repository at this point

• Fix CPU gather transposing column-contiguous slices (#3647)

  tillahoffmann

  authored
  b925592View commit details

  Copy full SHA for b925592

  Browse repository at this point

• Document default value of dilation in conv layer docstrings (#3811)

  Pablosinyores

  authored
  a956bcbView commit details

  Copy full SHA for a956bcb

  Browse repository at this point

• Fix wrong type parameter passed to gemm_splitk_nax (#3810)

  zcbenz

  authored
  a8c3e9cView commit details

  Copy full SHA for a8c3e9c

  Browse repository at this point

• Use large runner for Windows CUDA builds (#3809)

  zcbenz

  authored
  eba2b50View commit details

  Copy full SHA for eba2b50

  Browse repository at this point

• Fix compiled kernel correctness for negative-strided inputs (#3720)

  Show description for b5404c9
  

  3 peopleauthored
  b5404c9View commit details

  Copy full SHA for b5404c9

  Browse repository at this point

• Fix binary_cross_entropy docstring argument order (weights before with_logits) (#3807)

  Pablosinyores

  authored
  b837849View commit details

  Copy full SHA for b837849

  Browse repository at this point

• Add math mode option for custom Metal kernels (#3728)

  Show description for 51bef6f
  

  shubhxho

  and

  zcbenz

  authored
  51bef6fView commit details

  Copy full SHA for 51bef6f

  Browse repository at this point

• Fix Conv3d docstring argument order (padding before dilation) (#3793)

  Pablosinyores

  authored
  12ae9eaView commit details

  Copy full SHA for 12ae9ea

  Browse repository at this point

• Fix CUDA RMSNorm small-row dispatch (#3792)

  Show description for a5a684d
  

  dhiltgen

  and

  zcbenz

  authored
  a5a684dView commit details

  Copy full SHA for a5a684d

  Browse repository at this point

Commits on Jul 6, 2026

• Fix fp quantized matvec for output dim < 8 (#3804)

  Show description for 1700b39
  

  katlun-lgtm

  and

  katlun-lgtm

  authored
  1700b39View commit details

  Copy full SHA for 1700b39

  Browse repository at this point

• Bypass the PyTorch index TLS issues (#3808)

  angeloskath

  authored
  1c99a41View commit details

  Copy full SHA for 1c99a41

  Browse repository at this point

• Refactor macOS CI (#3806)

  zcbenz

  authored
  c018dc0View commit details

  Copy full SHA for c018dc0

  Browse repository at this point

Commits on Jul 3, 2026

• Fix HardShrink to accept its documented lambd argument (#3786)

  Pablosinyores

  authored
  de7b4edView commit details

  Copy full SHA for de7b4ed

  Browse repository at this point

• Fix kl_div_loss docstring formula and two doc nits (#3788)

  Pablosinyores

  authored
  34d465bView commit details

  Copy full SHA for 34d465b

  Browse repository at this point

Commits on Jul 1, 2026

• Fix BatchNorm docstring to cover 4D input and test the NHWC path (#3782)

  Pablosinyores

  authored
  e9463bbView commit details

  Copy full SHA for e9463bb

  Browse repository at this point

• fix: Quote hostname in mlx.launch ssh commands (#3783)

  varshneydevansh

  authored
  3e65a0fView commit details

  Copy full SHA for 3e65a0f

  Browse repository at this point

• Fix multi-wire recv prefill deadlock in jaccl ring backend (#3654)

  jasonpaulso

  authored
  61b5ff8View commit details

  Copy full SHA for 61b5ff8

  Browse repository at this point

Commits on Jun 29, 2026

• Add CI for Windows CUDA build (#3775)

  zcbenz

  authored
  25616a0View commit details

  Copy full SHA for 25616a0

  Browse repository at this point

Commits on Jun 28, 2026

• Fix Upsample align_corners singleton output (#3769)

  ishtihoss

  authored
  1d20ba3View commit details

  Copy full SHA for 1d20ba3

  Browse repository at this point

• Fix complex vjps for several unary ops (#3766)

  

  obchain

  authored
  af55406View commit details

  Copy full SHA for af55406

  Browse repository at this point

• array API: add cumulative_sum and cumulative_prod (#3731)

  Show description for eba3e4e
  

  katlun-lgtm

  and

  zcbenz

  authored
  eba3e4eView commit details

  Copy full SHA for eba3e4e

  Browse repository at this point

• array API: add positive, logical_xor, trunc, count_nonzero, diff, full_like (#3730)

  Show description for e94b415
  

  katlun-lgtm

  and

  zcbenz

  authored
  e94b415View commit details

  Copy full SHA for e94b415

  Browse repository at this point

• Enable fused SDPA vector kernel for asymmetric Q/V head dims (192, 128) (#3637)

  Show description for c9ccaba
  

  yohann-bearzi

  and

  zcbenz

  authored
  c9ccabaView commit details

  Copy full SHA for c9ccaba

  Browse repository at this point

• Fix CPU dynamic slice copy bound for collapsed shapes (#3739)

  Lyxot

  authored
  1bf65e3View commit details

  Copy full SHA for 1bf65e3

  Browse repository at this point

• Add vecdot to array API namespace (#3748)

  wgu9

  authored
  e37e926View commit details

  Copy full SHA for e37e926

  Browse repository at this point

• Fix logsumexp jvp to reduce along the axis (#3708)

  

  obchain

  authored
  da38f3eView commit details

  Copy full SHA for da38f3e

  Browse repository at this point

• Fix docstring nits (#3758)

  ishtihoss

  authored
  ca60290View commit details

  Copy full SHA for ca60290

  Browse repository at this point

Commits on Jun 26, 2026

• Add small-batch quantized matvec kernel (qmv_wide) (#3764)

  jessegross

  authored
  548dd80View commit details

  Copy full SHA for 548dd80

  Browse repository at this point

• [CUDA] [Improvement] Rope without copy (#3704)

  nastya236

  authored
  39886deView commit details

  Copy full SHA for 39886de

  Browse repository at this point

• Document row-wise sparsity in nn.init.sparse (#3752)

  Pablosinyores

  authored
  0386ad4View commit details

  Copy full SHA for 0386ad4

  Browse repository at this point

• metal: cache x in registers in rms_single_row to avoid redundant global read (#3754)

  will-march

  authored
  a131614View commit details

  Copy full SHA for a131614

  Browse repository at this point

Commits on Jun 24, 2026

• Fix build on Apple clang 15 in cpu_nd_array_to_mlx (#3751)

  Show description for b410f6c
  

  obchain

  and

  zcbenz

  authored
  b410f6cView commit details

  Copy full SHA for b410f6c

  Browse repository at this point