Cheng 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

..

mlx

Compatibility with pip-installed openmpi (#2741)

2025-11-07 16:58:31 -08:00

scripts

Linux on arm (#2751)

2025-11-11 11:41:14 -08:00

src

Fix dequantize python sig (dtype default) (#2752)

2025-11-11 09:55:24 -08:00

tests

[CUDA] cuDNN forward attention (#2743)

2025-11-14 09:23:56 +09:00