| .. |
|
composable_kernel @ 8182976c37
|
d8f104e97a
Support AMD ROCm on FlashAttention 2 (#1010)
|
4 月之前 |
|
cutlass @ 756c351b49
|
74b0761ff7
[FA3] BF16 forward
|
5 月之前 |
|
flash_attn
|
65f723bb9a
Split bwd into more .cu files to speed up compilation
|
4 月之前 |
|
flash_attn_ck
|
d8f104e97a
Support AMD ROCm on FlashAttention 2 (#1010)
|
4 月之前 |
|
ft_attention
|
50896ec574
Make nvcc threads configurable via environment variable (#885)
|
9 月之前 |
|
fused_dense_lib
|
50896ec574
Make nvcc threads configurable via environment variable (#885)
|
9 月之前 |
|
fused_softmax
|
50896ec574
Make nvcc threads configurable via environment variable (#885)
|
9 月之前 |
|
layer_norm
|
50896ec574
Make nvcc threads configurable via environment variable (#885)
|
9 月之前 |
|
rotary
|
50896ec574
Make nvcc threads configurable via environment variable (#885)
|
9 月之前 |
|
xentropy
|
50896ec574
Make nvcc threads configurable via environment variable (#885)
|
9 月之前 |
Tri Dao
65f723bb9a
Split bwd into more .cu files to speed up compilation