All issues

Search results

mx.dequantize returns garbage on CPU past 2^31 unpacked bits (int size in cpu/binary.h)

Status: Open.

#4479 In ml-explore/mlx;

· builtbycnob opened on Sep 8, 2026

mx.distributed ring backend: gradient-shaped CPU-stream all_sum deadlocks probabilistically (~3-4%/call) across a training loop; GPU-stream variant crashes deterministically at ~5s

Status: Open.

#4475 In ml-explore/mlx;

· ictashik opened on Sep 8, 2026

mx.compile cache has no eviction: one live Metal-buffer trace retained per distinct input shape, hits the 499000 buffer cap on variable-length training

Status: Open.

#4464 In ml-explore/mlx;

· chimera00 opened on Sep 4, 2026

Einsum selects unsupported matmul for integer contractions

documentation

low priority

Status: Open.

#4463 In ml-explore/mlx;

· justinchuby opened on Sep 4, 2026

Scan primitive lacks output_shapes for shapeless compile

enhancement

low priority

Status: Open.

#4460 In ml-explore/mlx;

· justinchuby opened on Sep 4, 2026

[BUG] Inconsistent behaviour of the floot_divide integer arrays truncates towards 0

bug

low priority

Status: Open.

#4450 In ml-explore/mlx;

· aaishwarymishra opened on Sep 2, 2026

[BUG]: bytes(array) dumps the wrong memory when the array is not packed

bug

low priority

python

Status: Open.

#4445 In ml-explore/mlx;

· simeetnayan81 opened on Sep 1, 2026

[BUG] Silently wrong grad and vmap for a strided slice that selects exactly one element

bug

low priority

Status: Open.

#4444 In ml-explore/mlx;

· mentria-ai opened on Sep 1, 2026

Metal: distributed collectives run on the CPU stream, and with the default fence each crossing costs ~180 µs (MLX_METAL_FAST_SYNCH: ~4 µs)

distributed

performance

Status: Open.

#4438 In ml-explore/mlx;

· avlp12 opened on Aug 31, 2026

nn.Upsample(mode="linear") is 2-4x slower than a matmul-based bilinear resize when upsampling

performance

Status: Open.

#4435 In ml-explore/mlx;

· bahaehmimdi opened on Aug 30, 2026

[ENH]: Using IndexError instead of ValueError at appropriate places?

enhancement

low priority

python

Status: Open.

#4428 In ml-explore/mlx;

· prady0t opened on Aug 29, 2026

[BUG] NAX MoE simdgroup skipping in 0.32.2 triggers lost JACCL completion under heterogeneous TP

bug

distributed

Status: Open.

#4403 In ml-explore/mlx;

· jonathan308 opened on Aug 25, 2026