mlx 0.31.2

[BrewTestBot]

Created by brew bump

---

Created with brew bump-formula-pr.

• resource blocks have been checked for updates.

release notes

## Highlights
* Wider support for cuda quantized matmuls (#3352, #3268, #3321, #3417, #3255)
* MLX can be used by multiple threads for independent computations (#3405, #3348, #3281, #3423)
* Added CUDA FFT support
* JACCL is now a standalone lib (#3412)
What's Changed

Bump by @angeloskath in Bump ml-explore/mlx#3244
win: re-enable and fix cuDNN performance by @dhiltgen in win: re-enable and fix cuDNN performance ml-explore/mlx#3242
Fix crashes in multi-threaded process teardown by @louen in Fix crashes in multi-threaded process teardown ml-explore/mlx#3167
[CUDA] Add FFT support by @lucasnewman in [CUDA] Add FFT support ml-explore/mlx#3243
[CUDA] Implement MaskedScatter  by @Lyxot in [CUDA] Implement MaskedScatter  ml-explore/mlx#3151
docs: fix PyTorch to MLX conversion example by @LxYuan0420 in docs: fix PyTorch to MLX conversion example ml-explore/mlx#3265
update requirements for Macbook Neo by @tosh in update requirements for Macbook Neo ml-explore/mlx#3257
fix comparison op JVP returning bool tangents instead of input dtype by @mm65x in fix comparison op JVP returning bool tangents instead of input dtype ml-explore/mlx#3253
fix nn.GRU skipping bhn bias when hidden is None by @mm65x in fix nn.GRU skipping bhn bias when hidden is None ml-explore/mlx#3252
[CUDA] Pipelined QMM by @zcbenz in [CUDA] Pipelined QMM ml-explore/mlx#3255
tests: harden memory leak check in test_siblings_without_eval by @booxter in tests: harden memory leak check in test_siblings_without_eval ml-explore/mlx#3088
Slice update with operation by @angeloskath in Slice update with operation ml-explore/mlx#3266
Nax Refactor by @jagrit06 in Nax Refactor ml-explore/mlx#3271
Fix building with CUDA toolkit 13.2 by @zcbenz in Fix building with CUDA toolkit 13.2 ml-explore/mlx#3273
[CUDA] fp and int4 quants for qmm_sm80 by @zcbenz in [CUDA] fp and int4 quants for qmm_sm80 ml-explore/mlx#3268
Fix repr of conv layers by @angeloskath in Fix repr of conv layers ml-explore/mlx#3275
Merge DeviceStream into CommandEncoder by @zcbenz in Merge DeviceStream into CommandEncoder ml-explore/mlx#3264
[CUDA] Search system-installed CUDA toolkit for headers by @zcbenz in [CUDA] Search system-installed CUDA toolkit for headers ml-explore/mlx#3277
Create default random key lazily by @zcbenz in Create default random key lazily ml-explore/mlx#3278
Support indexing with any type which implmented __index__ by @aisk in Support indexing with any type which implmented __index__ ml-explore/mlx#3210
Fix sort NaN handling for float16 and bfloat16 by @Lyxot in Fix sort NaN handling for float16 and bfloat16 ml-explore/mlx#3269
Use thread local storage for frontend compile cache by @zcbenz in Use thread local storage for frontend compile cache ml-explore/mlx#3280
[Metal][Performance]: Add split-K for quantized matmul (small M) by @Ziqiao-git in [Metal][Performance]: Add split-K for quantized matmul (small M) ml-explore/mlx#3120
[Metal] Fix depthwise conv 1D kernel name for large variant by @Brooooooklyn in [Metal] Fix depthwise conv 1D kernel name for large variant ml-explore/mlx#3289
Fix stale transform copy-chain leaks by @Brooooooklyn in Fix stale transform copy-chain leaks ml-explore/mlx#3290
Implement Pad::vmap to replace NYI stub by @Aristide021 in Implement Pad::vmap to replace NYI stub ml-explore/mlx#3304
logo files by @andresy in logo files ml-explore/mlx#3308
Fix vmap + floor_divide: preserve integer dtype by @robert-johansson in Fix vmap + floor_divide: preserve integer dtype ml-explore/mlx#3292
Fix moved-from shape bug in broadcast_arrays causing vmap bus error by @Aristide021 in Fix moved-from shape bug in broadcast_arrays causing vmap bus error ml-explore/mlx#3310
Use nb::ndarray for checking arrays by @zcbenz in Use nb::ndarray for checking arrays ml-explore/mlx#3283
Add output_shapes for AddMM by @pHequals7 in Add output_shapes for AddMM ml-explore/mlx#3262
Manage Metal objects with smart pointers by @zcbenz in Manage Metal objects with smart pointers ml-explore/mlx#3282
[CUDA] support sorting complex numbers by @Lyxot in [CUDA] support sorting complex numbers ml-explore/mlx#3286
Add norm parameter to FFT transforms (backward/ortho/forward) by @Aristide021 in Add norm parameter to FFT transforms (backward/ortho/forward) ml-explore/mlx#3287
Make each thread have its own default stream by @zcbenz in Make each thread have its own default stream ml-explore/mlx#3281
[CUDA] Implement BlockMaskedMM by @Lyxot in [CUDA] Implement BlockMaskedMM ml-explore/mlx#3299
Fix np bfloat16 misinterpreted as complex by @kellen-sun in Fix np bfloat16 misinterpreted as complex ml-explore/mlx#3146
Remove no longer needed const_cast by @zcbenz in Remove no longer needed const_cast ml-explore/mlx#3325
Bump actions/deploy-pages from 4 to 5 by @dependabot[bot] in Bump actions/deploy-pages from 4 to 5 ml-explore/mlx#3334
Fix use after move by @angeloskath in Fix use after move ml-explore/mlx#3343
Decouple CommandEncoder from Device by @zcbenz in Decouple CommandEncoder from Device ml-explore/mlx#3316
Add vmap for BroadcastAxes by @angeloskath in Add vmap for BroadcastAxes ml-explore/mlx#3344
Add fftfreq, rfftfreq and scalar axes for fftshift/ifftshift by @declanhealy2 in Add fftfreq, rfftfreq and scalar axes for fftshift/ifftshift ml-explore/mlx#3298
[Metal] Support sorting complex numbers by @Lyxot in [Metal] Support sorting complex numbers ml-explore/mlx#3314
[CUDA] Fallback QMM by @zcbenz in [CUDA] Fallback QMM ml-explore/mlx#3315
Make CommandEncoder thread local by @zcbenz in Make CommandEncoder thread local ml-explore/mlx#3348
[CUDA] 3/5/6-bit quants for qmm_naive by @zcbenz in [CUDA] 3/5/6-bit quants for qmm_naive ml-explore/mlx#3352
Fix regression in array creation by @angeloskath in Fix regression in array creation ml-explore/mlx#3353
Use metal as the front-end for the metal linker by @louen in Use metal as the front-end for the metal linker ml-explore/mlx#3354
Add printoptions by @ChristophePRAT in Add printoptions ml-explore/mlx#3333
Add a convenience for making local streams in python by @angeloskath in Add a convenience for making local streams in python ml-explore/mlx#3355
Fix CMake finding wrong Python during pip install by @fijimunkii in Fix CMake finding wrong Python during pip install ml-explore/mlx#3375
[CUDA] Add GatherQMM for quantized gather matmul by @Lyxot in [CUDA] Add GatherQMM for quantized gather matmul ml-explore/mlx#3321
fix: fail build when Metal compiler header resolution fails by @dogukanveziroglu in fix: fail build when Metal compiler header resolution fails ml-explore/mlx#3332
Fix: Correct cross-attention query routing in Post-LN TransformerDecoderLayer by @suryawanshishantanu6 in Fix: Correct cross-attention query routing in Post-LN TransformerDecoderLayer ml-explore/mlx#3382
[CUDA] Thread safety by @zcbenz in [CUDA] Thread safety ml-explore/mlx#3367
Fix test "test get streams" missing initialization by @dseredkin in Fix test "test get streams" missing initialization ml-explore/mlx#3376
Conjugate VJP and JVP support by @CameronChurchwell in Conjugate VJP and JVP support ml-explore/mlx#3386
Fix int16 overflow in SDPA NAX mask indexing for KV sequences > 32K by @Clydingus in Fix int16 overflow in SDPA NAX mask indexing for KV sequences > 32K ml-explore/mlx#3361
Avoid joining threads on exit by @zcbenz in Avoid joining threads on exit ml-explore/mlx#3388
Add clear_streams API for cleanup before exit by @zcbenz in Add clear_streams API for cleanup before exit ml-explore/mlx#3395
Update nanobind version to v2.12.0 by @jrp2014 in Update nanobind version to v2.12.0 ml-explore/mlx#3396
Jaccl refactor by @angeloskath in Jaccl refactor ml-explore/mlx#3412
Fixes for CUDA CI by @zcbenz in Fixes for CUDA CI ml-explore/mlx#3413
Validate safetensors data offsets by @MillaFleurs in Validate safetensors data offsets ml-explore/mlx#3364
Validate safetensors data offsets against file boundaries by @matinsaurralde in Validate safetensors data offsets against file boundaries ml-explore/mlx#3410
Document sort stability and NaN handling by @NeuralNoble in Document sort stability and NaN handling ml-explore/mlx#3400
ThreadLocalStream in C++ by @zcbenz in ThreadLocalStream in C++ ml-explore/mlx#3405
Fix jaccl init bug by @angeloskath in Fix jaccl init bug ml-explore/mlx#3418
Segmented mm nax kernel by @angeloskath in Segmented mm nax kernel ml-explore/mlx#3419
[CUDA] gather_mm by @zcbenz in [CUDA] gather_mm ml-explore/mlx#3414
[CUDA] GatherQMM matrix-matrix sm80/naive path  by @Lyxot in [CUDA] GatherQMM matrix-matrix sm80/naive path  ml-explore/mlx#3417
[CUDA] Handle residue k in qmm_naive by @zcbenz in [CUDA] Handle residue k in qmm_naive ml-explore/mlx#3379
Speed up NAX split-K by better tuning and routing and fix NAX addmm by @angeloskath in Speed up NAX split-K by better tuning and routing and fix NAX addmm ml-explore/mlx#3422
Make Scheduler::enqueue thread safe by @zcbenz in Make Scheduler::enqueue thread safe ml-explore/mlx#3423
Fix flaky TestVmap.test_vmap_masked_scatter by @zcbenz in Fix flaky TestVmap.test_vmap_masked_scatter ml-explore/mlx#3421
Fix synchronize for ThreadLocalStream by @angeloskath in Fix synchronize for ThreadLocalStream ml-explore/mlx#3429
Fix bytes_per_key truncation in random kernels (Metal + CUDA) by @dogukanveziroglu in Fix bytes_per_key truncation in random kernels (Metal + CUDA) ml-explore/mlx#3432
Throw meaningful error when Metal device is not found by @dogukanveziroglu in Throw meaningful error when Metal device is not found ml-explore/mlx#3428
Fix kernel cache collision in Compiled constructor by @dogukanveziroglu in Fix kernel cache collision in Compiled constructor ml-explore/mlx#3427
Fix mx.prod vjp for complex types by @CameronChurchwell in Fix mx.prod vjp for complex types ml-explore/mlx#3433

New Contributors

@LxYuan0420 made their first contribution in docs: fix PyTorch to MLX conversion example ml-explore/mlx#3265
@tosh made their first contribution in update requirements for Macbook Neo ml-explore/mlx#3257
@mm65x made their first contribution in fix comparison op JVP returning bool tangents instead of input dtype ml-explore/mlx#3253
@booxter made their first contribution in tests: harden memory leak check in test_siblings_without_eval ml-explore/mlx#3088
@Ziqiao-git made their first contribution in [Metal][Performance]: Add split-K for quantized matmul (small M) ml-explore/mlx#3120
@Brooooooklyn made their first contribution in [Metal] Fix depthwise conv 1D kernel name for large variant ml-explore/mlx#3289
@Aristide021 made their first contribution in Implement Pad::vmap to replace NYI stub ml-explore/mlx#3304
@pHequals7 made their first contribution in Add output_shapes for AddMM ml-explore/mlx#3262
@declanhealy2 made their first contribution in Add fftfreq, rfftfreq and scalar axes for fftshift/ifftshift ml-explore/mlx#3298
@fijimunkii made their first contribution in Fix CMake finding wrong Python during pip install ml-explore/mlx#3375
@dogukanveziroglu made their first contribution in fix: fail build when Metal compiler header resolution fails ml-explore/mlx#3332
@suryawanshishantanu6 made their first contribution in Fix: Correct cross-attention query routing in Post-LN TransformerDecoderLayer ml-explore/mlx#3382
@dseredkin made their first contribution in Fix test "test get streams" missing initialization ml-explore/mlx#3376
@CameronChurchwell made their first contribution in Conjugate VJP and JVP support ml-explore/mlx#3386
@Clydingus made their first contribution in Fix int16 overflow in SDPA NAX mask indexing for KV sequences > 32K ml-explore/mlx#3361
@jrp2014 made their first contribution in Update nanobind version to v2.12.0 ml-explore/mlx#3396
@matinsaurralde made their first contribution in Validate safetensors data offsets against file boundaries ml-explore/mlx#3410
@NeuralNoble made their first contribution in Document sort stability and NaN handling ml-explore/mlx#3400

Full Changelog: ml-explore/mlx@v0.31.1...v0.31.2

View the full release notes at https://github.com/ml-explore/mlx/releases/tag/v0.31.2.

---

[botantony]

  Files with missing rpath:
    /opt/homebrew/Cellar/mlx/0.31.2/lib/libmlx.dylib