ml-explore/mlx

View on GitHub

MLX: An array framework for Apple silicon

Stars: 28.5k
Forks: 2.3k
Open beginner issues: 0
Indexed issues: 98
Avg merge: 3d 8h
Merged PRs (30d): 62

Dominant language: C++
License: MIT
Last GitHub push: Sep 17, 2026
Latest indexed: Sep 19, 2026
Contributing guide: Contributing guide
Code of conduct: Code of conduct
Beginner labels: No beginner labels indexed

Beginner friendly Everything

Open All

98 open issues indexed

`mlx` no longer builds from source with Xcode 27 (macOS 27.0 SDK) Open

Difficulty 2/5 1-3 hours Newbie friendliness 88/100

ml-explore/mlx#4533 · Opened Sep 18, 2026
[BUG] Metal `col_reduce_longcolumn` produces wrong results for negative-stride views Open

Difficulty 3/5 1-2 days Newbie friendliness 25/100

ml-explore/mlx#4528 · Opened Sep 17, 2026
ParallelFileReader: uncatchable fatal error when pread lands at EOF Open

Difficulty 4/5 3-5 days Newbie friendliness 58/100

ml-explore/mlx#4526 · Opened Sep 17, 2026
macOS 26 wheel is built for 26.2 but tagged macosx_26_0 — pip installs it on 26.0/26.1, where NAX is enabled on M5-class GPUs Open

Difficulty 4/5 3-5 days Newbie friendliness 55/100

ml-explore/mlx#4525 · Opened Sep 17, 2026
Metal command-buffer limits (50 ops / 50 MB) cost 5–8% decode throughput on launch-heavy models (M3 Ultra measurements) Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

ml-explore/mlx#4521 · Opened Sep 16, 2026
[Performance] gather_qmm at M=1 streams 4-bit expert weights at ~half the bandwidth of M=4 (MoE decode) Open

Difficulty 5/5 Over a week Newbie friendliness 45/100

ml-explore/mlx#4509 · Opened Sep 14, 2026
[BUG] threads and streams issues Open

Difficulty 5/5 Over a week Newbie friendliness 30/100

ml-explore/mlx#4506 · 1 comment · Opened Sep 14, 2026
mx.compile inlines float scalar constants with 7 significant digits, so compiled results differ from eager by 1 ulp Open

Difficulty 2/5 1-3 hours Newbie friendliness 85/100

ml-explore/mlx#4503 · 1 comment · Opened Sep 14, 2026
[BUG] `argsort` diverges between cpu and gpu on complexes with NaN Open

Difficulty 3/5 1-2 days Newbie friendliness 68/100

ml-explore/mlx#4502 · Opened Sep 14, 2026
quantized_matmul is slower than dense bf16 when the output dimension is not a multiple of 8, and ~7x slower below 8 Open

Difficulty 4/5 3-5 days Newbie friendliness 42/100

ml-explore/mlx#4498 · Opened Sep 13, 2026
[BUG] JACCL segfaults in mx.distributed.init(backend="jaccl") on 2-node Mac-Mac with RDMA devices present Open

Difficulty 4/5 3-5 days Newbie friendliness 45/100

ml-explore/mlx#4485 · 1 comment · Opened Sep 9, 2026
mx.dequantize returns garbage on CPU past 2^31 unpacked bits (int size in cpu/binary.h) Open

bug low priority

Difficulty 4/5 3-5 days Newbie friendliness 54/100

ml-explore/mlx#4479 · 4 comments · Opened 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 Open

bug distributed

Difficulty 5/5 Over a week Newbie friendliness 35/100

ml-explore/mlx#4475 · 1 comment · Opened Sep 8, 2026
Einsum selects unsupported matmul for integer contractions Open

documentation low priority

Difficulty 4/5 3-5 days Newbie friendliness 68/100

ml-explore/mlx#4463 · 5 comments · Opened Sep 4, 2026
Scan primitive lacks output_shapes for shapeless compile Open

enhancement low priority

Difficulty 3/5 1-2 days Newbie friendliness 78/100

ml-explore/mlx#4460 · Opened Sep 4, 2026
[BUG] Inconsistent behaviour of the floot_divide integer arrays truncates towards 0 Open

bug low priority

Difficulty 3/5 1-2 days Newbie friendliness 68/100

ml-explore/mlx#4450 · 5 comments · Opened Sep 2, 2026
[BUG] Silently wrong grad and vmap for a strided slice that selects exactly one element Open

bug low priority

Difficulty 3/5 1-2 days Newbie friendliness 76/100

ml-explore/mlx#4444 · Opened 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) Open

distributed performance

Difficulty 5/5 Over a week Newbie friendliness 35/100

ml-explore/mlx#4438 · 2 comments · Opened Aug 31, 2026
`nn.Upsample(mode="linear")` is 2-4x slower than a matmul-based bilinear resize when upsampling Open

performance

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ml-explore/mlx#4435 · 4 comments · Opened Aug 30, 2026
[ENH]: Using `IndexError` instead of `ValueError` at appropriate places? Open

enhancement low priority python

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ml-explore/mlx#4428 · 3 comments · Opened Aug 29, 2026
[BUG] NAX MoE simdgroup skipping in 0.32.2 triggers lost JACCL completion under heterogeneous TP Open

bug distributed

Difficulty 5/5 Over a week Newbie friendliness 38/100

ml-explore/mlx#4403 · 1 comment · Opened Aug 25, 2026
[BUG] Mixed slice + fancy indexing produces wrong shape/garbage data for out-of-range slice bounds Open

bug low priority

Difficulty 3/5 1-2 days Newbie friendliness 68/100

ml-explore/mlx#4399 · Opened Aug 25, 2026
[BUG] mx.var / mx.std wrong for offset data Open

bug low priority

Difficulty 4/5 3-5 days Newbie friendliness 55/100

ml-explore/mlx#4379 · 3 comments · Opened Aug 23, 2026
[BUG] `__setitem__` with a single Ellipsis fails Open

bug low priority python

Difficulty 3/5 1-2 days Newbie friendliness 65/100

ml-explore/mlx#4358 · Opened Aug 19, 2026
Support fusing element-wise operations into reductions Open

enhancement

Difficulty 4/5 3-5 days Newbie friendliness 45/100

ml-explore/mlx#4355 · 2 comments · Opened Aug 19, 2026
[CUDA] gather_qmm ~24x slower than equal-FLOPs quantized_matmul on GB10 (sm_121) — makes quantized MoE prefill unusable Open

cuda performance

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ml-explore/mlx#4339 · 2 comments · Opened Aug 18, 2026
vmap of scatter silently returns wrong values when the vmap axis is >= 2 Open

bug low priority

Difficulty 2/5 1-3 hours Newbie friendliness 72/100

ml-explore/mlx#4321 · 1 comment · Opened Aug 17, 2026
[BUG] JACCL SIGSEGV in tbt_post_recv during unique all_sum (no unplug, no model, ports still up) Open

bug distributed

Difficulty 4/5 3-5 days Newbie friendliness 38/100

ml-explore/mlx#4319 · 1 comment · Opened Aug 17, 2026
[BUG] JACCL never detects a lost peer: survivors busy-wait at 100% CPU indefinitely Open

bug distributed low priority

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ml-explore/mlx#4278 · 1 comment · Opened Aug 16, 2026
Ring backend: recv fails with errno 14 (EFAULT) during multi-node pipeline inference, fabric-independent (distinct from #3862) Open

bug distributed low priority

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ml-explore/mlx#4276 · 1 comment · Opened Aug 15, 2026
Would new fused ops be welcome in mx.fast? Open

question

Difficulty 5/5 Over a week Newbie friendliness 48/100

ml-explore/mlx#4259 · 2 comments · Opened Aug 14, 2026
[BUG] JACCL segfaults in tbt_post_recv when a Thunderbolt cable is unplugged mid-collective Open

bug distributed

Difficulty 4/5 3-5 days Newbie friendliness 45/100

ml-explore/mlx#4192 · 1 comment · Opened Aug 11, 2026
[BUG] `mx.distributed.all_max` / `all_min` silently drop `NaN` (inconsistent with `mx.max`/`mx.min`) Open

bug distributed low priority

Difficulty 3/5 1-2 days Newbie friendliness 68/100

ml-explore/mlx#4072 · 4 comments · Opened Aug 8, 2026
[New feature] Support exporting memory snapshots compatible with PyTorch's memory viz format Open

enhancement low priority

Difficulty 5/5 Over a week Newbie friendliness 25/100

ml-explore/mlx#3954 · 2 comments · Opened Jul 31, 2026
Memory leak when compiled function contains multi-output op Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ml-explore/mlx#3932 · 5 comments · Opened Jul 27, 2026
CMake cannot build Metal kernels for iOS Open

enhancement

Difficulty 4/5 3-5 days Newbie friendliness 45/100

ml-explore/mlx#3915 · 1 comment · Opened Jul 25, 2026
[BUG] JACCL MeshImpl::recv spins forever on peer loss — silent hang, no timeout (UC) Open

bug distributed

Difficulty 5/5 Over a week Newbie friendliness 28/100

ml-explore/mlx#3910 · 2 comments · Opened Jul 24, 2026
[CUDA] GLM-5.2/DeepseekV32 mxfp4 forward numerically wrong on Blackwell (sm_120/121/110), correct on Metal — individual ops fine, composed forward diverges ~25% Open

bug cuda

ml-explore/mlx#3879 · 1 comment · 1 assignee · Opened Jul 21, 2026
[CUDA] distributed all_sum barrier hangs in cu::AtomicEvent::wait on Blackwell (sm_121/sm_110) during warmup — ring sockets healthy Open

bug cuda distributed

Difficulty 4/5 3-5 days Newbie friendliness 35/100

ml-explore/mlx#3876 · 3 comments · Opened Jul 20, 2026
matmul ~36× slower than cuBLAS on Windows when data are allocated in host Open

bug cuda low priority

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ml-explore/mlx#3861 · 5 comments · Opened Jul 17, 2026
mx.sum over a non-last axis returns all zeros for certain large tensors (e.g. [1, 32, 4096, 131072]) Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 45/100

ml-explore/mlx#3784 · 4 comments · Opened Jun 30, 2026
[BUG] Complex vjps are wrong for cos, sqrt, and rsqrt (missing conjugate) Open

Difficulty 3/5 1-2 days Newbie friendliness 68/100

ml-explore/mlx#3780 · Opened Jun 29, 2026
[BUG] Complex vjps are wrong for arcsin/arccos/arctan/arcsinh/arccosh/arctanh (missing conjugate) Open

Difficulty 3/5 1-2 days Newbie friendliness 70/100

ml-explore/mlx#3778 · Opened Jun 29, 2026
[Feature]: Supporting normal python datatypes in the grad method obtained from mlx.core.value_and_grad Open

enhancement low priority

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ml-explore/mlx#3774 · 2 comments · Opened Jun 26, 2026
Feature Request: Least Squares Solver Open

enhancement low priority

Difficulty 5/5 Over a week Newbie friendliness 38/100

ml-explore/mlx#3773 · 2 comments · Opened Jun 26, 2026
[BUG] mx.expm1 is wrong for complex inputs (discards imaginary part) Open

bug

Difficulty 2/5 1-3 hours Newbie friendliness 72/100

ml-explore/mlx#3759 · Opened Jun 24, 2026
[BUG] ring and jaccl both fail to connect (error 60/65) on 4-node M3 Ultra cluster — IP network verified routable, full mesh confirmed Open

bug distributed low priority

Difficulty 5/5 Over a week Newbie friendliness 38/100

ml-explore/mlx#3755 · 2 comments · Opened Jun 23, 2026
[BUG] jvp of mx.logcumsumexp is not implemented Open

enhancement

Difficulty 3/5 1-2 days Newbie friendliness 67/100

ml-explore/mlx#3710 · Opened Jun 17, 2026
[BUG] MLX doesn't publish iOS-compatible wheels Open

enhancement low priority python

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ml-explore/mlx#3665 · 1 comment · 1 reaction · Opened Jun 12, 2026
[BUG] crash on qr followed by diag Open

bug

Difficulty 4/5 3-5 days Newbie friendliness 48/100

ml-explore/mlx#3583 · 1 comment · Opened May 23, 2026

ml-explore/mlx

Get new issues in your inbox