mlx 0.30.0

[BrewTestBot]

Created by brew bump

---

Created with brew bump-formula-pr.

• resource blocks have been checked for updates.

release notes

## Highlights

Support for Neural Accelerators on M5 (macOS >= 26.2)

What's Changed

Fix AdamW weight_decay default value in docstring by @goingreen in Fix AdamW weight_decay default value in docstring ml-explore/mlx#2557
Fix dequantize python sig by @wrmsr in Fix dequantize python sig ml-explore/mlx#2562
fix copies in sdpa by @awni in fix copies in sdpa ml-explore/mlx#2563
chore: Update Docs With Slice Copy Example by @krishi-saripalli in chore: Update Docs With Slice Copy Example ml-explore/mlx#2559
Fixed several type annotations in the MLX stubs which degraded to Unknown/Any by @Maalvi14 in Fixed several type annotations in the MLX stubs which degraded to Unknown/Any ml-explore/mlx#2560
typing: add type hints to mlx.core.array, linalg, and random by @XXXXRT666 in typing: add type hints to mlx.core.array, linalg, and random ml-explore/mlx#2565
Set ccache size before building by @zcbenz in Set ccache size before building ml-explore/mlx#2570
Faster fully depthwise-separable 1D conv by @awni in Faster fully depthwise-separable 1D conv ml-explore/mlx#2567
Fix a few ccache cache miss by @zcbenz in Fix a few ccache cache miss ml-explore/mlx#2573
Some tweaks in cmake files by @zcbenz in Some tweaks in cmake files ml-explore/mlx#2574
Add batch offsets for mx.fast.rope by @awni in Add batch offsets for mx.fast.rope ml-explore/mlx#2564
[CUDA] Use GEMM with epilogue instead of AddMM by @zcbenz in [CUDA] Use GEMM with epilogue instead of AddMM ml-explore/mlx#2569
[CUDA] Fix alpha not respected when using bias epilogue by @zcbenz in [CUDA] Fix alpha not respected when using bias epilogue ml-explore/mlx#2578
Fix flaky addmm tests by @zcbenz in Fix flaky addmm tests ml-explore/mlx#2581
Adding Relu2 by @Goekdeniz-Guelmez in Adding Relu2 ml-explore/mlx#2582
Add sdpa with sinks by @awni in Add sdpa with sinks ml-explore/mlx#2558
[CUDA] Set bias as input when using bias epilogue by @zcbenz in [CUDA] Set bias as input when using bias epilogue ml-explore/mlx#2584
[CUDA] Fix NCCL stub for release build by @awni in [CUDA] Fix NCCL stub for release build ml-explore/mlx#2587
patch bump by @awni in patch bump ml-explore/mlx#2588
Refactor code examples to use 'gelu' by @umbertomig in Refactor code examples to use 'gelu' ml-explore/mlx#2592
Fix metal scan by @awni in Fix metal scan ml-explore/mlx#2591
Fix typo in average_gradients function call by @umbertomig in Fix typo in average_gradients function call ml-explore/mlx#2594
No copy batch rope by @awni in No copy batch rope ml-explore/mlx#2595
Update export function example for array input by @umbertomig in Update export function example for array input ml-explore/mlx#2598
Expose mx.depends to Python by @awni in Expose mx.depends to Python ml-explore/mlx#2606
fix: library loading for swift dynamic frameworks by @bilousoleksandr in fix: library loading for swift dynamic frameworks ml-explore/mlx#2568
Detect cache thrashing in LRUCache by @zcbenz in Detect cache thrashing in LRUCache ml-explore/mlx#2600
Lower sorted QMM gather threshold by @awni in Lower sorted QMM gather threshold ml-explore/mlx#2609
implement Convolution::output_shape by @josharian in implement Convolution::output_shape ml-explore/mlx#2601
Avoid producing NaN in attention by @awni in Avoid producing NaN in attention ml-explore/mlx#2608
[CUDA] Recycle CUDA events by @zcbenz in [CUDA] Recycle CUDA events ml-explore/mlx#2604
[CUDA] fix cudaGraphLaunch by @CC-Yeh in [CUDA] fix cudaGraphLaunch ml-explore/mlx#2613
Support pickling array for bfloat16 by @CC-Yeh in Support pickling array for bfloat16 ml-explore/mlx#2586
New tuning for small K gemv by @jagrit06 in New tuning for small K gemv ml-explore/mlx#2620
Allow None input to compiled functions by @awni in Allow None input to compiled functions ml-explore/mlx#2621
Compiled should not end in broadcast by @angeloskath in Compiled should not end in broadcast ml-explore/mlx#2622
Bump the version by @angeloskath in Bump the version ml-explore/mlx#2627
[CUDA] Make CudaEvent work with multi-device by @zcbenz in [CUDA] Make CudaEvent work with multi-device ml-explore/mlx#2614
Fix incorrect path and typos by @aisk in Fix incorrect path and typos ml-explore/mlx#2630
Fix for max block dim by @awni in Fix for max block dim ml-explore/mlx#2631
Compile now can attach arbitrary data to an entry by @angeloskath in Compile now can attach arbitrary data to an entry ml-explore/mlx#2634
[CUDA] Wait for tasks in cuda by @awni in [CUDA] Wait for tasks in cuda ml-explore/mlx#2636
Fix status message by @angeloskath in Fix status message ml-explore/mlx#2638
fix cross entropy axis param by @awni in fix cross entropy axis param ml-explore/mlx#2641
Faster triu, tril, where with scalar by @awni in Faster triu, tril, where with scalar ml-explore/mlx#2644
[CUDA] Add a small column specialization to reduce by @angeloskath in [CUDA] Add a small column specialization to reduce ml-explore/mlx#2642
[CUDA] Fix flaky test by @awni in [CUDA] Fix flaky test ml-explore/mlx#2646
Configure CMake to export compile_commands.json by @andportnoy in Configure CMake to export compile_commands.json ml-explore/mlx#2645
Faster complex matmul by @CC-Yeh in Faster complex matmul ml-explore/mlx#2571
Fix compile when outputs change by @awni in Fix compile when outputs change ml-explore/mlx#2648
Speed up compile for node with many parents by @awni in Speed up compile for node with many parents ml-explore/mlx#2649
Fix and refactor row-reduce by @angeloskath in Fix and refactor row-reduce ml-explore/mlx#2650
[CUDA] Fix jit file cache for large kernel names by @angeloskath in [CUDA] Fix jit file cache for large kernel names ml-explore/mlx#2656
Fix all_gather vjp by @awni in Fix all_gather vjp ml-explore/mlx#2654
Fix fast synch when fence is waited before a command buffer is created by @awni in Fix fast synch when fence is waited before a command buffer is created ml-explore/mlx#2657
Fix cumulative operations when axis=None by @aisk in Fix cumulative operations when axis=None ml-explore/mlx#2653
Export with callback by @awni in Export with callback ml-explore/mlx#2612
bump patch by @awni in bump patch ml-explore/mlx#2658
Enable addmm low-precision cpu by @awni in Enable addmm low-precision cpu ml-explore/mlx#2661
Precise sigmoid by @awni in Precise sigmoid ml-explore/mlx#2659
Debug cuda conv by @awni in Debug cuda conv ml-explore/mlx#2662
Speed up scalars part 2 by @awni in Speed up scalars part 2 ml-explore/mlx#2669
Normalize README bullet formatting and other Markdown small fixes by @Mistobaan in Normalize README bullet formatting and other Markdown small fixes ml-explore/mlx#2671
Modified sort behavior when running CPU or Metal to match NumPy/JAX by @Maalvi14 in Modified sort behavior when running CPU or Metal to match NumPy/JAX ml-explore/mlx#2667
remove unused unary file by @awni in remove unused unary file ml-explore/mlx#2672
Nccl timeout by @nastya236 in Nccl timeout ml-explore/mlx#2673
suppress gcc 10.1 warnings by @awni in suppress gcc 10.1 warnings ml-explore/mlx#2679
patch bump by @awni in patch bump ml-explore/mlx#2680
Improved mx.split() docs by @Maalvi14 in Improved mx.split() docs ml-explore/mlx#2689
fix warnings showing up with -Wall by @andresy in fix warnings showing up with -Wall ml-explore/mlx#2692
Einsum error msg improvement by @Maalvi14 in Einsum error msg improvement ml-explore/mlx#2690
optionally load metallib from framework by @davidkoski in optionally load metallib from framework ml-explore/mlx#2702
Fix addmm cpu for beta != 1.0 by @awni in Fix addmm cpu for beta != 1.0 ml-explore/mlx#2699
Add mx.median op by @awni in Add mx.median op ml-explore/mlx#2705
bump python by @awni in bump python ml-explore/mlx#2694
Fp8 conversion by @awni in Fp8 conversion ml-explore/mlx#2686
fix: linux-{fedora}x86_64-build by @incertum in fix: linux-{fedora}x86_64-build ml-explore/mlx#2707
Add quantize/dequantize for mxfp8 and nvfp4 by @awni in Add quantize/dequantize for mxfp8 and nvfp4 ml-explore/mlx#2688
Migrate CircleCI to GitHub Actions by @madrob in Migrate CircleCI to GitHub Actions ml-explore/mlx#2716
Fix KeyError for missing domain_uuid_key in Thunderbolt setup by @thechriswebb in Fix KeyError for missing domain_uuid_key in Thunderbolt setup ml-explore/mlx#2682
fix memory count bug by @awni in fix memory count bug ml-explore/mlx#2717
Fix the order of hosts in the ring by @angeloskath in Fix the order of hosts in the ring ml-explore/mlx#2718
Fix docs path by @madrob in Fix docs path ml-explore/mlx#2719
Use faster dequant for fp4 by @awni in Use faster dequant for fp4 ml-explore/mlx#2720
update: add linux fedora container CI - CPP build test only by @incertum in update: add linux fedora container CI - CPP build test only ml-explore/mlx#2722
add null check -- the bundleIdentifier is optional by @davidkoski in add null check -- the bundleIdentifier is optional ml-explore/mlx#2709
Fix compile multi capture by @awni in Fix compile multi capture ml-explore/mlx#2678
Set up publishing to PyPI and Test-PyPI by @madrob in Set up publishing to PyPI and Test-PyPI ml-explore/mlx#2721
Check isnan in maximum / minimum with CPU backend by @aisk in Check isnan in maximum / minimum with CPU backend ml-explore/mlx#2652
Fix addmm with empty matrices and beta != 1.0 by @harsh-sutariya in Fix addmm with empty matrices and beta != 1.0 ml-explore/mlx#2715
skip self-hosted runners on forks by @madrob in skip self-hosted runners on forks ml-explore/mlx#2730
only build for macos 14 and up by @awni in only build for macos 14 and up ml-explore/mlx#2731
don't test when doing release by @awni in don't test when doing release ml-explore/mlx#2734
Make cpu binary_op easily accessible by @angeloskath in Make cpu binary_op easily accessible ml-explore/mlx#2733
fix property name by @madrob in fix property name ml-explore/mlx#2736
Nccl reduce scatter, all gather by @nastya236 in Nccl reduce scatter, all gather ml-explore/mlx#2727
[CUDA] Reduce use of managed memory by @awni in [CUDA] Reduce use of managed memory ml-explore/mlx#2725
Shapeless support for zeros/ones_like by @CC-Yeh in Shapeless support for zeros/ones_like ml-explore/mlx#2726
Compatibility with pip-installed openmpi by @pcuenca in Compatibility with pip-installed openmpi ml-explore/mlx#2741
Fix release builds by @awni in Fix release builds ml-explore/mlx#2746
patch bump by @awni in patch bump ml-explore/mlx#2750
Fix dequantize python sig (dtype default) by @wrmsr in Fix dequantize python sig (dtype default) ml-explore/mlx#2752
remove circle by @awni in remove circle ml-explore/mlx#2753
Fix irregular_strides benchmark shape type by @wrmsr in Fix irregular_strides benchmark shape type ml-explore/mlx#2754
Linux on arm by @awni in Linux on arm ml-explore/mlx#2751
minor debugging for publishing by @madrob in minor debugging for publishing ml-explore/mlx#2739
Export custom kernel by @awni in Export custom kernel ml-explore/mlx#2756
Fix slice with negative strides by @awni in Fix slice with negative strides ml-explore/mlx#2758
[CUDA] Check CUDA error in synchronize by @zcbenz in [CUDA] Check CUDA error in synchronize ml-explore/mlx#2757
fix release by @awni in fix release ml-explore/mlx#2759
[CUDA] cuDNN forward attention by @zcbenz in [CUDA] cuDNN forward attention ml-explore/mlx#2743
Fix exporting with constants by @awni in Fix exporting with constants ml-explore/mlx#2769
Separate test-linux from build-linux/cuda in GitHub Actions by @zcbenz in Separate test-linux from build-linux/cuda in GitHub Actions ml-explore/mlx#2765
[CUDA] Use arch specific targets when possible by @awni in [CUDA] Use arch specific targets when possible ml-explore/mlx#2771
Fix MPI distributed tests with CUDA backend by @zcbenz in Fix MPI distributed tests with CUDA backend ml-explore/mlx#2775
Fix warnings with cmake 4.1 by @zcbenz in Fix warnings with cmake 4.1 ml-explore/mlx#2774
Use ccache in GitHub Actions by @zcbenz in Use ccache in GitHub Actions ml-explore/mlx#2773
[CUDA] Tune ops per buffer based on device by @awni in [CUDA] Tune ops per buffer based on device ml-explore/mlx#2761
fix release 2 by @awni in fix release 2 ml-explore/mlx#2767
Run CI for pushes by @zcbenz in Run CI for pushes ml-explore/mlx#2777
Remove pip cache in GitHub Actions by @zcbenz in Remove pip cache in GitHub Actions ml-explore/mlx#2776
Build and test with multiple CUDA versions by @zcbenz in Build and test with multiple CUDA versions ml-explore/mlx#2780
Use std::optional for mask_arr arg by @zcbenz in Use std::optional for mask_arr arg ml-explore/mlx#2763
Do not run CPU tests in CUDA builds by @zcbenz in Do not run CPU tests in CUDA builds ml-explore/mlx#2784
Test every commit in main branch by @zcbenz in Test every commit in main branch ml-explore/mlx#2781
Fix nightly build by @zcbenz in Fix nightly build ml-explore/mlx#2785
Remove unneeded tests in nightly build by @zcbenz in Remove unneeded tests in nightly build ml-explore/mlx#2786
Fix building with CUDA < 12.8 by @zcbenz in Fix building with CUDA < 12.8 ml-explore/mlx#2782
Avoid duplicate CI runs when starting a PR from upstream branch by @zcbenz in Avoid duplicate CI runs when starting a PR from upstream branch ml-explore/mlx#2788
build docs on linux by @awni in build docs on linux ml-explore/mlx#2787
[CUDA] cuDNN backward attention by @zcbenz in [CUDA] cuDNN backward attention ml-explore/mlx#2762
more accurate rope fallback by @awni in more accurate rope fallback ml-explore/mlx#2792
Fix version tag by @awni in Fix version tag ml-explore/mlx#2790
version by @awni in version ml-explore/mlx#2797
Add Masked Scatter by @CC-Yeh in Add Masked Scatter ml-explore/mlx#2663
Add Neural Accelerator Support by @jagrit06 in Add Neural Accelerator Support ml-explore/mlx#2772

New Contributors

@goingreen made their first contribution in Fix AdamW weight_decay default value in docstring ml-explore/mlx#2557
@krishi-saripalli made their first contribution in chore: Update Docs With Slice Copy Example ml-explore/mlx#2559
@Maalvi14 made their first contribution in Fixed several type annotations in the MLX stubs which degraded to Unknown/Any ml-explore/mlx#2560
@XXXXRT666 made their first contribution in typing: add type hints to mlx.core.array, linalg, and random ml-explore/mlx#2565
@umbertomig made their first contribution in Refactor code examples to use 'gelu' ml-explore/mlx#2592
@bilousoleksandr made their first contribution in fix: library loading for swift dynamic frameworks ml-explore/mlx#2568
@josharian made their first contribution in implement Convolution::output_shape ml-explore/mlx#2601
@aisk made their first contribution in Fix incorrect path and typos ml-explore/mlx#2630
@Mistobaan made their first contribution in Normalize README bullet formatting and other Markdown small fixes ml-explore/mlx#2671
@incertum made their first contribution in fix: linux-{fedora}x86_64-build ml-explore/mlx#2707
@thechriswebb made their first contribution in Fix KeyError for missing domain_uuid_key in Thunderbolt setup ml-explore/mlx#2682
@harsh-sutariya made their first contribution in Fix addmm with empty matrices and beta != 1.0 ml-explore/mlx#2715
@pcuenca made their first contribution in Compatibility with pip-installed openmpi ml-explore/mlx#2741

Full Changelog: ml-explore/mlx@v0.29.0...v0.30.0

View the full release notes at https://github.com/ml-explore/mlx/releases/tag/v0.30.0.

---

[p-linnane]

We need the Metal toolchain on our CI images for this. I'll take care of that tomorrow.

[p-linnane]

Building for x86_64 arch is not officially supported.

[botantony]

Good job @p-linnane!

[github-actions]

🤖 An automated task has requested bottles to be published to this PR.

Caution

Please do not push to this PR branch before the bottle commits have been pushed, as this results in a state that is difficult to recover from. If you need to resolve a merge conflict, please use a merge commit. Do not force-push to this PR branch.