dan_the_3rd 9b8166e3f0 fMHA: Add backward pass (#844) ...

* fMHA: Add backward pass

* Better checks for strides/alignments

* Remove fb-internal URL

* torch.Tensor.untyped_storage requires pytorch 2.0+

* minor changes

* make test

---------

Co-authored-by: danthe3rd <danthe3rd>
Co-authored-by: Haicheng Wu <haichengw@nvidia.com>

2023-04-06 20:44:58 -04:00

..

custom_mma_base.h

New updates for 2.11 (#775)

2023-01-20 16:32:57 -05:00

custom_mma_multistage.h

New updates for 2.11 (#775)

2023-01-20 16:32:57 -05:00

custom_mma_pipelined.h

Fix typos 2 (#842)

2023-03-09 23:22:56 -05:00

custom_mma.h

New updates for 2.11 (#775)

2023-01-20 16:32:57 -05:00

find_default_mma.h

fMHA: Sync FW with xFormers (#828)

2023-02-22 23:25:31 -05:00

mma_accum_lambda_iterator.h

fMHA: Sync FW with xFormers (#828)

2023-02-22 23:25:31 -05:00

mma_from_smem.h

fMHA: Add backward pass (#844)

2023-04-06 20:44:58 -04:00