Awni Hannun
b61a65e313
fix copies in sdpa ( #2563 )
2025-09-02 11:00:36 -07:00
Awni Hannun
8ce49cd39e
fix quantized vjp for mxfp4 ( #2555 )
2025-08-29 10:06:15 -07:00
Awni Hannun
70560b6bd5
Add mode parameter for quantization ( #2499 )
...
* add mode parameter for quantization
* mxfp4 quantize/dequantize + start of optional biases
* mxfp4 works
* speedup
* cpu mxfp4
* fix
* fix test tol
* fix
* refactor
* add quant mode enum
2025-08-28 06:45:26 -07:00
Awni Hannun
7ef8a6f2d5
[CUDA] fix sort ( #2550 )
...
* [CUDA] fix sort
* fix test
2025-08-27 19:48:43 -07:00
Awni Hannun
5458d43247
add load with path tests ( #2543 )
2025-08-26 14:24:47 -07:00
Awni Hannun
3dcb286baf
Remove stream from average grads so it uses default ( #2532 )
...
* Remove stream from average grads so it uses default
* comment
2025-08-25 15:56:29 -07:00
Cheng
4822c3dbe9
[CUDA] Implement DynamicSlice/DynamicSliceUpdate ( #2533 )
...
* Move DynamicSlice to gpu/primitives
* Implement compute_dynamic_offset in CUDA
2025-08-26 07:31:39 +09:00
Anastasiia Filippova
9392fc3f88
NCCL backend ( #2476 )
2025-08-21 11:56:15 -07:00
Awni Hannun
e843c4d8d5
fix power ( #2523 )
2025-08-21 06:46:01 -07:00
Angelos Katharopoulos
e397177f6e
Custom cuda kernel ( #2517 )
2025-08-20 17:20:22 -07:00
Cheng
f4c8888cbe
[CUDA] Fix stride of singleton dims before passing to cuDNN ( #2521 )
2025-08-21 08:55:26 +09:00
Angelos Katharopoulos
25c1e03205
Fix overflow in large filter small channels ( #2520 )
2025-08-20 08:03:29 -07:00
Cheng
ac85ddfdb7
[CUDA] Add GEMM-based fallback convolution kernels ( #2511 )
...
* Add gemm_conv
* Add gemm_grouped_conv
2025-08-20 10:06:22 +09:00
Awni Hannun
e7c6e1db82
no segfault with uninitialized array.at ( #2514 )
2025-08-18 08:33:38 -07:00
Awni Hannun
c5fcd5b61b
fix custom kernel test ( #2510 )
2025-08-18 06:45:59 -07:00
Cheng
1ba18ff7d9
[CUDA] Fix conv grads with groups ( #2495 )
...
* Put reshape utils in one file
* [CUDA] Fix conv grads with groups
* Put the reshape utils in gpu/copy.h
2025-08-16 10:09:18 +09:00
Luca Vivona
728d4db582
Support destination arg in tree flatten/unflatten ( #2450 )
2025-08-06 15:34:59 -07:00
Awni Hannun
fa89f0b150
faster gather qmm sorted test ( #2463 )
2025-08-05 06:27:40 -07:00
Awni Hannun
0b807893a7
fix wraps compile ( #2461 )
2025-08-04 16:14:18 -07:00
Cheng
86c6a15571
[CUDA] Backward convolution ( #2431 )
2025-08-01 09:54:05 +09:00
junpeiz
8b25ce62d5
Add tests for export including control flow models and quantized models ( #2430 )
...
* Add tests for export, including control flow export and quantized model export.
* Skip quantization related test for CUDA backend.
2025-07-31 11:06:26 -07:00
Awni Hannun
d32519c8ee
fix gemv regression ( #2445 )
2025-07-30 14:23:01 -07:00
Awni Hannun
b405591249
fix circular reference ( #2443 )
2025-07-30 09:37:44 -07:00
Awni Hannun
ef631d63af
faster rms norm ( #2433 )
2025-07-29 13:12:00 -07:00
Awni Hannun
5597fa089c
Fix qvm splitk ( #2415 )
2025-07-25 11:50:24 -07:00
Cheng
6f5874a2f2
[CUDA] Initial implementation of Convolution with cuDNN ( #2385 )
...
* Link with cuDNN
* Initial implementation
* Remove backend apis
* Fix recording cudnn conv
* More unused backend apis
* Fix C++ conv tests
* include cudnn as python dep
* Install libcudnn9-dev-cuda-12 in CI
* cudnn only accepts contiguous inputs
* Switch to backend apis
* Plan needs to be kept alive
* Turn off tf32
* Add cache
* Test the native cuda graph api
* Set cudnn stream before execution
* Make LRUCache more like a normal container
* Do error check for cublas handle
* Zero-initilizing array
* Use tf32 for conv
* Skip TestConv.test_torch_conv_2D test
---------
Co-authored-by: Awni Hannun <awni@apple.com >
2025-07-25 08:12:10 +09:00
Awni Hannun
e1840853ce
full row mask in sdpa consistently gives nan ( #2406 )
2025-07-23 16:37:03 -07:00
Gökdeniz Gülmez
deee214a95
Adding support for the Muon Optimizer ( #1914 )
...
* initial commit with workong optmimizer
* update ACKNOWLEDGMENTS.md
* nits and adding it to test
* nits
* G.astype(mx.bfloat16) to G.astype(G.dtype)
* G.ndim >= 2 to assert G.ndim == 2
* remove coments
* replace with mx.addmm
* remove comments
* format
* nits
* match muon
* fix addmm
---------
Co-authored-by: Awni Hannun <awni@apple.com >
2025-07-18 12:25:28 -07:00
Awni Hannun
f409b229a4
fix ring distributed test ( #2380 )
2025-07-16 11:25:24 -07:00
Awni Hannun
d7734edd9f
fix complex reduce + nan propagation in min and max ( #2377 )
2025-07-15 18:19:47 -07:00
Awni Hannun
49114f28ab
fix flaky test ( #2371 )
2025-07-14 17:16:18 -07:00
Awni Hannun
e7d2ebadd2
[CUDA] Affine quantize ( #2354 )
...
* affine quantize and dequantize kernels
* format
* fix
* format
2025-07-14 15:45:44 -07:00
Angelos Katharopoulos
5201df5030
Fix imag() vjp ( #2367 )
2025-07-14 13:11:16 -07:00
Cheng
8347575ba1
[CUDA] Implement Scan kernel ( #2347 )
...
* Contiguous scan
* Strided scan
* Enable tests
* Fix failing logaddexp test
* Use cexpf in Metal
2025-07-10 16:54:12 -07:00
Angelos Katharopoulos
0eb035b4b1
Fix type promotion in Adam with bias correction ( #2350 )
2025-07-10 11:14:42 -07:00
jhavukainen
8c7bc30ce4
Align mlx::core::min op nan propagation with NumPy ( #2346 )
2025-07-10 06:20:43 -07:00
jhavukainen
8b9a3f3cea
Align mlx::core::max op nan propagation with NumPy ( #2339 )
...
* Make max op NaN propagation rules align with numpy
* Adding benchmarks and testing for max op nanpropagation
* Pre-commit formatting
* Fix max complex64 nan propagation and add test
* Improve the cpp unittest
* Only check nans on non-integral types in simd_reduce_impl.
* Cleanup using namespace alias
* Add cpu Max nanpropagation. Fix a small fib in cpu max dispatch data types for int8/int16.
* Make the max nanpropagation test more meaningful for integer types
* Remove tuple unpacking syntax to comply with earlier python versions. Add cuda skip to nanpropagation tests, fix cuda implementation in a separate PR.
2025-07-09 11:26:27 -07:00
Angelos Katharopoulos
4a9b29a875
MoE backward improvements ( #2335 )
2025-07-07 17:59:53 -07:00
Awni Hannun
ec0d5db67b
[CUDA] Switch to CUDA graphs ( #2317 )
...
* cuda graph prototype
fix signal bug + start to add dependencies
capture more
capture more ops
remaining ops
fix reduce and rope deps
add concurrent context
try update, but not working
cosistent topology order
use node api
use node api directly to reduce overhead
fix bug
use kernels in unary
cache graph
format
fix synchronization
format
* comment
2025-07-02 15:59:13 -07:00
Awni Hannun
cfb6a244ea
allow parameters to be deleted ( #2325 )
2025-07-01 21:27:23 -07:00
Awni Hannun
dd4f53db63
use fp32 for testing, add more complex ops ( #2322 )
2025-07-01 07:30:00 -07:00
Awni Hannun
33bf1a244b
Fix module update in strict mode ( #2321 )
...
* fix module update in strict mode
* allow GELU to be pickled
2025-06-29 11:12:29 -07:00
Angelos Katharopoulos
772f471ff2
[CUDA] Fix reductions ( #2314 )
2025-06-27 12:59:20 -07:00
Angelos Katharopoulos
2c11d10f8d
Split broadcast so it is always fused in compile ( #2318 )
2025-06-26 22:08:18 -07:00
Awni Hannun
81bb9a2a9e
Compile float64 functions on CPU ( #2311 )
2025-06-24 10:18:52 -07:00
Angelos Katharopoulos
5adf185f86
Fix update_modules() when providing a subset ( #2308 )
2025-06-20 17:19:46 -07:00
Awni Hannun
cad5c0241c
[CUDA] synch properly waits for all tasks to finish and clear ( #2303 )
...
* cuda synch properly waits for all tasks to finish and clear
* fix copy
2025-06-17 12:03:25 -07:00
Awni Hannun
b8022c578a
divmod, partition, sort fixes ( #2302 )
2025-06-16 18:49:32 -07:00
Awni Hannun
bc53f8293f
Cuda bug fixes 2 ( #2298 )
...
* more bug fixes
* more bug fixes
* format
2025-06-16 13:14:46 -07:00
Awni Hannun
c552ff2451
[CUDA] Fix back-end bugs and enable corresponding tests ( #2296 )
...
* Fix some cuda back-end bugs and enable corresponding tests
* more fixes
* enable more tests
* format
2025-06-16 08:45:40 -07:00
