All issues

Search results

Expose precompiled_metal_kernel to load offline-compiled .metallib (-Ofast) for custom kernels

Status: Open.

#4541 In ml-explore/mlx;

· dahai80 opened on Sep 21, 2026

[BUG] mx.max() with slice returns incorrect values both on CPU and GPU when the element is more than 2^31

Status: Open.

#4539 In ml-explore/mlx;

· kvcache670 opened on Sep 21, 2026

[BUG] mlx.sort return value incorrect on GPU

Status: Open.

#4538 In ml-explore/mlx;

· kvcache670 opened on Sep 20, 2026

Grouped Conv1d / Conv2d initialization uses ungrouped fan-in

Status: Open.

#4536 In ml-explore/mlx;

· Evihut opened on Sep 19, 2026

[BUG] fast.metal_kernel silently truncates the Metal dispatch: group_dims clamped against grid dimensions in CustomKernel::eval_gpu

Status: Open.

#4534 In ml-explore/mlx;

· pseudobacon opened on Sep 19, 2026

mlx no longer builds from source with Xcode 27 (macOS 27.0 SDK)

Status: Open.

#4533 In ml-explore/mlx;

· jrp2014 opened on Sep 18, 2026

[BUG] Metal col_reduce_longcolumn produces wrong results for negative-stride views

Status: Open.

#4528 In ml-explore/mlx;

· mateuuszzzzz opened on Sep 17, 2026

ParallelFileReader: uncatchable fatal error when pread lands at EOF

Status: Open.

#4526 In ml-explore/mlx;

· pseudobacon opened on Sep 17, 2026

macOS 26 wheel is built for 26.2 but tagged macosx_26_0 — pip installs it on 26.0/26.1, where NAX is enabled on M5-class GPUs

Status: Open.

#4525 In ml-explore/mlx;

· ChanLumerico opened on Sep 17, 2026

Metal command-buffer limits (50 ops / 50 MB) cost 5–8% decode throughput on launch-heavy models (M3 Ultra measurements)

Status: Open.

#4521 In ml-explore/mlx;

· freddyhaddad opened on Sep 16, 2026

[Performance] gather_qmm at M=1 streams 4-bit expert weights at ~half the bandwidth of M=4 (MoE decode)

Status: Open.

#4509 In ml-explore/mlx;

· freddyhaddad opened on Sep 14, 2026

[BUG] threads and streams issues

Status: Open.

#4506 In ml-explore/mlx;

· davidkoski opened on Sep 14, 2026