Search results

[BUG] mlx-0.31.2+mlx-metal-0.31.2 works for inference, but locally built mlx-0.32.0.dev20260524+2165dc08 produces gibberish. Metal compiler regression?

Status: Open.

#3586 In ml-explore/mlx;

· jrp2014 opened on May 24, 2026

mx.fast.scaled_dot_product_attention produces incorrect attention scores (MLX 0.31.2)

Status: Open.

#3585 In ml-explore/mlx;

· coldbugs 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

bug

Status: Open.

#3573 In ml-explore/mlx;

· JacobLinCool opened on May 21, 2026

mx.random.normal: bit-exact divergence on Apple M1 Max vs M3 Ultra / M5

bug

Status: Open.

#3568 In ml-explore/mlx;

· electron-rare opened on May 20, 2026

JIT compilation for CUTLASS/CuTe kernels

cuda

enhancement

Status: Open.

#3567 In ml-explore/mlx;

· zcbenz opened on May 20, 2026

[BUG] Metal GPU crash (SIGABRT) on Apple M4 Max with mlx_lm.server running Qwen3.5-27B-4bit

Status: Open.

#3564 In ml-explore/mlx;

· halubzbm-creator opened on May 19, 2026

[Performance] qmv kernel: non-linear cost step at M=3 for large MLP shapes

performance

Status: Open.

#3553 In ml-explore/mlx;

· AirRunner opened on May 16, 2026

Slow 1D dot product mx.inner compared to custom kernel

performance

Status: Open.

#3533 In ml-explore/mlx;

· SuriyaaMM opened on May 12, 2026

Add full array-api tests

enhancement

python

Status: Open.

#3514 In ml-explore/mlx;

· ilan-gold opened on May 10, 2026

Why is MLX_BUILD_METAL disabled on iOS

enhancement

good first issue

Status: Open.

#3493 In ml-explore/mlx;

· olokobayusuf opened on May 8, 2026