Search results

[BUG] JVPs of power, divmod, and slice_update mishandle partly traced inputs

Status: Open.

#3634 In ml-explore/mlx;

· qflen opened on Jun 6, 2026

[BUG] JVP non-deterministic crashes when called in a loop.

bug

Status: Open.

#3629 In ml-explore/mlx;

· kyrollosyanny opened on Jun 4, 2026

[BUG] JVP silently returns a zero tangent when differentiating through mx.where

bug

Status: Open.

#3627 In ml-explore/mlx;

· kyrollosyanny opened on Jun 4, 2026

[Performance] conv_general 5D (3D convolution) is 2-5x slower than equivalent per-frame 2D conv

performance

Status: Open.

#3625 In ml-explore/mlx;

· kimjj81 opened on Jun 4, 2026

compile: failed first trace leaves a stale cache entry that returns empty-success on retry

bug

Status: Open.

#3624 In ml-explore/mlx;

· al8n opened on Jun 4, 2026

[BUG] gemma-4-12b with omlx error loading model

Status: Open.

#3623 In ml-explore/mlx;

· benb0jangles opened on Jun 4, 2026

Data race: process-global InTracing::trace_stack_ static is not thread_local; concurrent compile/grad/vmap traces race other threads

bug

Status: Open.

#3620 In ml-explore/mlx;

· al8n opened on Jun 4, 2026

mx.fast.metal_kernel: Add support for compiler options (-fmetal-math-mode, integer template parameters, Metal 4 Tensor types

enhancement

Status: Open.

#3592 In ml-explore/mlx;

· rajveer43 opened on May 27, 2026

Xcode 26.5 / metalfe-32023.883 mis-compiles NAX GEMM kernel

bug

Status: Open.

#3586 In ml-explore/mlx;

· jrp2014 opened on May 24, 2026

qmm_splitk dispatch causes 1.5-1.8x regression vs fp16 at M=96, M=128 on NAX hardware

performance

Status: Open.

#3584 In ml-explore/mlx;

· marian-gg opened on May 23, 2026

[BUG] crash on qr followed by diag

bug

Status: Open.

#3583 In ml-explore/mlx;

· hughjonesd opened on May 23, 2026

[BUG] mx.fast.scaled_dot_product_attention: position-0 output changes at K=8 → K=9, all dtypes

enhancement

Status: Open.

#3573 In ml-explore/mlx;

· JacobLinCool opened on May 21, 2026