| .. |
|
fmha
|
Support all head dims that are multiples of 8, up to 128
|
2022-10-24 16:04:21 -07:00 |
|
.DS_Store
|
Rename, add benchmarking script
|
2022-05-26 13:57:38 -07:00 |
|
fmha_block_dgrad_fp16_kernel_loop.sm80.cu
|
Implement cross attention
|
2022-07-03 17:48:12 -07:00 |
|
fmha_block_dgrad_kernel_1xN_loop.h
|
Support all head dims that are multiples of 8, up to 128
|
2022-10-24 16:04:21 -07:00 |
|
fmha_block_fprop_fp16_kernel.sm80.cu
|
Implement cross attention
|
2022-07-03 17:48:12 -07:00 |
|
fmha_block_fprop_kernel_1xN.h
|
Support all head dims that are multiples of 8, up to 128
|
2022-10-24 16:04:21 -07:00 |
|
fmha_blockmask.h
|
Implement cross attention
|
2022-07-03 17:48:12 -07:00 |
|
fmha_dgrad_fp16_kernel_loop.sm80.cu
|
Support all head dims that are multiples of 8, up to 128
|
2022-10-24 16:04:21 -07:00 |
|
fmha_dgrad_kernel_1xN_loop.h
|
Support all head dims that are multiples of 8, up to 128
|
2022-10-24 16:04:21 -07:00 |
|
fmha_fprop_fp16_kernel.sm80.cu
|
Support all head dims that are multiples of 8, up to 128
|
2022-10-24 16:04:21 -07:00 |
|
fmha_fprop_kernel_1xN.h
|
Get rid of o_rows_are_valid since we don't have headdim=16 anymore
|
2022-10-24 17:29:36 -07:00 |
|
fmha_kernel.h
|
Implement cross attention
|
2022-07-03 17:48:12 -07:00 |
|
fmha_utils.h
|
Implement for bf16
|
2022-07-09 23:31:56 -07:00 |
|
fmha.h
|
Split bwd on the seqlen_q dimension
|
2022-10-23 11:35:15 -07:00 |
|
fp16_switch.h
|
Fixed switch statement, thanks @yocabon
|
2022-10-04 21:31:39 -04:00 |
|
philox.cuh
|
Rework dropout to decouple forward and backward
|
2022-10-21 12:04:27 -07:00 |
|
static_switch.h
|
Implement attention kernel that splits the batch into two
|
2022-10-13 20:49:02 -07:00 |
