Search results

[Metal] NVFP4: 3-tier global_scale unsupported + 2-tier simdgroup-lane indexing bug

Status: Open.

#3557 In ml-explore/mlx;

· yohann-bearzi opened on May 16, 2026

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

Status: Open.

#3553 In ml-explore/mlx;

· AirRunner opened on May 16, 2026

Metal: QQMatmul::eval_gpu gemv path silently drops global_scale_x/global_scale_w for nvfp4

Status: Open.

#3550 In ml-explore/mlx;

· Brooooooklyn opened on May 14, 2026

[ops] Add mx.empty() — uninitialized array allocation

Status: Open.

#3549 In ml-explore/mlx;

· megacpp opened on May 13, 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

mx.compile rejects custom cache objects — no pytree registration mechanism

enhancement

python

Status: Open.

#3499 In ml-explore/mlx;

· st-adam opened on May 8, 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

MLXNN.RoPE produces row-asymmetric output for byte-identical row inputs at [B>1, H, 1, D]

bug

Status: Open.

#3496 In ml-explore/mlx;

· ivan-digital opened on May 7, 2026

cuda pypi wheel missing?

cuda

Status: Open.

#3489 In ml-explore/mlx;

· Evanev7 opened on May 6, 2026

[BUG] Missing array-api methods

enhancement

good first issue

python

Status: Open.

#3484 In ml-explore/mlx;

· ilan-gold opened on May 5, 2026

[BUG] JACCL "Changing queue pair to RTR failed with errno 22" on Apple Thunderbolt RDMA — GID selection regression in #3412

bug

distributed

Status: Open.

#3467 In ml-explore/mlx;

· danielkristofik opened on Apr 29, 2026