Issues · ml-explore/mlx · GitHub

Why not implement this in Pytorch?
#12 · ToniRV opened on Dec 6, 2023
29
Common Installation Problems
#200 · awni opened on Dec 17, 2023
11

Labels Milestones New issue

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

#3499

· st-adam opened

on May 8, 2026

[BUG] mx.fast.rope produces wrong output for batch rows >= 1 when seq_len == 1

#3494

· nicoveraz opened

on May 8, 2026

Why is `MLX_BUILD_METAL` disabled on iOS

#3493

· olokobayusuf opened

on May 8, 2026

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

#3496

· ivan-digital opened

on May 7, 2026

cuda pypi wheel missing?

#3489

· Evanev7 opened

on May 6, 2026

[BUG][CUDA] Illegal memory access in QMM when M/N is not multiple of tile size

good first issue

#3487

· zcbenz opened

on May 6, 2026

[BUG] Missing `array-api` methods

good first issue

#3484

· ilan-gold opened

on May 5, 2026

[BUG] `quantized_matmul` produces wrong results for GQA `expand_dims` broadcasting when M < `vector_limit`

#3480

· AirRunner opened

on May 4, 2026

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

#3467

· danielkristofik opened

on Apr 29, 2026

init(backend="any") selects ring singleton instead of JACCL when launched via mlx.launch --backend jaccl

#3442

· machiabeli opened

on Apr 22, 2026

[BUG] mlx-cpu backend has a hard dependncy on a deprecated version of gcc

#3430

· Evanev7 opened

on Apr 20, 2026

FSDP saving state improvement

#3401

· nastya236 opened

on Apr 12, 2026