2469fc2939
patch bump for next release ( #2991 )
2026-01-14 08:46:09 -08:00
099dcc0f4c
Expose to/from fp8 in Python and don't auto-convert fp8 when loading from safetensors ( #2985 )
2026-01-13 15:48:21 -08:00
4160ec10f7
Fix RandomBits::is_equivalent to include width ( #2978 )
...
Co-authored-by: KD2YCU <me@kd2ycu.com >
Co-authored-by: Angelos Katharopoulos <katharas@gmail.com >
Co-authored-by: Awni Hannun <awni@apple.com >
2026-01-13 12:42:37 -08:00
edab937248
Add asarray to __array_namespace__ ( #2966 )
2026-01-12 06:16:27 -08:00
1680b6fe38
fix numpy dtype bug ( #2960 )
2026-01-05 11:20:40 -08:00
1d21d0e696
[CUDA] Implement gather_mm_rhs ( #2902 )
2025-12-24 09:42:56 +09:00
1eef1d155c
Metal/CPU nvfp4 and mxfp8 ( #2946 )
2025-12-22 20:45:19 -08:00
c0fea26ed2
Fix for non row-contig scales ( #2941 )
2025-12-21 06:12:41 -08:00
81dfe5f137
Fix grad in place updates ( #2899 )
2025-12-11 14:44:58 -08:00
3c8ce9b00e
Fix input buffer donation in compile ( #2897 )
2025-12-11 06:07:03 -08:00
997cfc7699
Add a 2-pass col reduce for CUDA ( #2863 )
2025-12-04 15:53:59 -08:00
a6d6717181
fix compile copying ( #2871 )
2025-12-04 12:32:56 -08:00
941cfe23d7
Layer norm throws on dimension mismatch ( #2870 )
2025-12-04 11:21:05 -08:00
9abb0b8123
Added support for pytree types that inherit from tuple and typing.namedtuple ( #2845 )
2025-12-04 11:06:45 -08:00
cacbdbf995
Fix init from double ( #2861 )
2025-12-03 06:08:11 -08:00
d8ceae7b77
Reduce JVP ( #2854 )
2025-12-02 16:17:47 -08:00
eff0e31f00
Fix export scatters ( #2852 )
2025-12-02 11:24:40 -08:00
8879ee00eb
Support more Numpy interfaces for masked_scatter ( #2832 )
2025-12-01 17:51:02 -08:00
704fd1ae28
[CUDA] Support array mask in SDPA ( #2822 )
2025-11-26 11:08:58 +09:00
618c87af8c
Add float64 Eig and complex64 SVD/Eig support ( Fixes #2708 ) ( #2737 )
...
Co-authored-by: Awni Hannun <awni.hannun@gmail.com >
Co-authored-by: Awni Hannun <awni@apple.com >
2025-11-22 06:51:36 -08:00
4a09264236
Tolerance for some ops tests on cuda ( #2815 )
2025-11-21 16:06:16 -08:00
54f1cc6e3e
Add Neural Accelerator Support ( #2772 )
2025-11-19 15:06:00 -08:00
b3825ac149
Add Masked Scatter ( #2663 )
...
Co-authored-by: Awni Hannun <awni@apple.com >
Co-authored-by: Angelos Katharopoulos <katharas@gmail.com >
Co-authored-by: Angelos Katharopoulos <a_katharopoulos@apple.com >
2025-11-19 14:53:32 -08:00
f46877bc08
more accurate rope fallback ( #2792 )
2025-11-19 06:07:21 -08:00
6f35017d1b
[CUDA] cuDNN backward attention ( #2762 )
2025-11-19 08:13:50 +09:00
27ff069175
Fix exporting with constants ( #2769 )
2025-11-14 12:52:08 -08:00
3b2ffcefc3
[CUDA] cuDNN forward attention ( #2743 )
...
* Separate sdpa kernels in another file
* Initial support for cuDNN SDPA
* Diable a few corner cases
* Remove scaled_dot_product_attention.h
* Use cuDNN attention for prefilling
* cuDNN SDPA requires Ampere and later
* Address reviews
* Do contiguous copy of inputs
2025-11-14 09:23:56 +09:00
66519fb348
fix slice ( #2758 )
2025-11-13 11:30:02 -08:00
8973550ff3
export custom kernel ( #2756 )
2025-11-13 11:29:50 -08:00
be9e2aebd6
Shapeless support for zeros/ones_like ( #2726 )
...
* shapeless support for zeros/ones_like
* Improvements
* fix access after moved
2025-11-06 19:12:20 -08:00
27778156dc
Nccl reduce scatter, all gather ( #2727 )
...
* Added reduce scatter and all gather for nccl
* fix unused import, delete unused file
* small fix
* deleted useless condition
* fixed comments
* fix bug in eval_gpu, renamed to sum_scatter, fix docs
* final fix docs
* remove and
* Update mlx/distributed/mpi/mpi.cpp
Co-authored-by: Awni Hannun <awni.hannun@gmail.com >
* fix broken set input output
* fixes set output
* typo
* fix typo
* no cpu, no gpu for reduce scatter
---------
Co-authored-by: Awni Hannun <awni.hannun@gmail.com >
2025-11-05 08:21:11 -08:00
50fa315d18
Fix addmm with empty matrices and beta != 1.0 ( #2715 )
2025-11-03 14:16:15 -08:00
93d76b0f30
Fix compile multi capture ( #2678 )
...
* fix compile when compiling multiple lambdas with the same capture
* add test
2025-11-03 06:33:43 -08:00
68c5fa1c95
fix memory count bug ( #2717 )
2025-10-30 14:27:15 -07:00
ec72b44417
Add quantize/dequantize for mxfp8 and nvfp4 ( #2688 )
...
* Add quantize/dequantize slow path for mxfp8 and nvfp4
* fast cuda kernel for mx/nv quantization
* fallback for cuda < 12.8 (#2697 )
* format (#2700 )
* fix (#2701 )
* metal kernels
* docs
* fix jit
* add default bits and group sizes
* improve quant docs
* fix output type of mxfp4 matmuls
2025-10-28 16:23:12 -07:00
539d8322d1
add median op ( #2705 )
2025-10-27 11:33:42 -07:00
c4767d110f
fix addmm cpu ( #2699 )
2025-10-27 11:33:32 -07:00
9cbb1b0148
Modified sort behavior when running CPU or Metal to match NumPy/JAX ( #2667 )
...
* Modified sort behavior when running CPU or Metal to match NumPy/JAX sorting behavior.
* Modified sort behavior when running CPU or Metal to match NumPy/JAX
* nits
---------
Co-authored-by: Awni Hannun <awni@apple.com >
2025-10-13 14:36:45 -07:00
630350ad3e
Precise sigmoid ( #2659 )
...
* bump patch
* Sigmoid matches PyTorch and is more precise on tails
2025-10-10 10:05:23 -07:00
380aeb58ae
enable admm low-precision cpu ( #2661 )
2025-10-10 09:50:54 -07:00
e89e8b4272
Export with callback ( #2612 )
...
* export with callback
* export with callback
* Add types, fix kwarg ordering bug + test
* cleanup, test, fix
* typos
2025-10-08 19:24:33 -07:00
343e33b6d5
fix all_gather vjp ( #2654 )
2025-10-07 06:05:23 -07:00
a7a94b29d7
Fix compile when outputs change ( #2648 )
2025-10-03 08:40:57 -07:00
1c9ae1eaa1
cuda fix flaky test ( #2646 )
2025-10-02 15:40:04 -07:00
e88f2d4a8e
fix cross entropy axis param ( #2641 )
...
* fix cross entropy axis param
* faster grad clipping
2025-10-01 16:49:55 -07:00
eb24267b56
Compile now can attach arbitrary data to an entry ( #2634 )
2025-09-30 13:33:27 -07:00
dc371ae7a5
fix for max block dim ( #2631 )
2025-09-29 08:59:25 -07:00
d4f4ff3c5e
Allow None input to compiled functions ( #2621 )
...
* Allow None input to compiled functions
* Allow None input to compiled functions
2025-09-25 08:42:23 -07:00
fbbf3b9b3e
Support pickling array for bfloat16 ( #2586 )
...
* add bfloat16 pickling
* Improvements
* improve
---------
Co-authored-by: Chen-Chen Yeh <ge96noj@mytum.de >
2025-09-22 20:12:15 -07:00
711a645807
avoid producing NaN in attention ( #2608 )
2025-09-22 13:10:43 -07:00
Awni Hannun