issues
search
ROCm
/
flash-attention
Fast and memory-efficient exact attention
BSD 3-Clause "New" or "Revised" License
107
stars
33
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Use same python as build flash-attn to generate ck kernel
#66
oraluben
opened
1 week ago
0
Integrate ck tile backward
#65
rocking5566
closed
4 days ago
0
[Issue]: memory format option is only supported by strided tensors
#64
hyattpd
opened
1 week ago
0
Enable fwd and varlen_fwd on AMD
#63
micmelesse
closed
1 week ago
0
Feature/seqlenq ngroups swap
#62
poyenc
closed
1 week ago
0
Ck tile/flash attention
#61
rocking5566
closed
2 weeks ago
0
Enable fwd and varlen_fwd on AMD
#60
micmelesse
closed
1 week ago
1
Do not return lse if return_softmax == false
#59
rocking5566
closed
1 month ago
0
Fix stride issues in flash_attn_interface
#58
clintg6
opened
1 month ago
0
Integrate with ck
#57
rocking5566
closed
1 month ago
0
Integrate ck with FA
#56
rocking5566
closed
1 month ago
0
update from upstream
#55
rocking5566
closed
1 month ago
0
[Issue]: Expected dout_seq_stride == out_seq_stride to be true, but got false
#54
ehartford
closed
1 month ago
2
[Feature]: Support for newer flash-attention versions (e.g. ≥2.1.0)
#53
JiahuaZhao
opened
1 month ago
2
GPUAI-1250 - Flash Attention v2.04 two modules layer_norm cannot be used fixed
#52
xiaoxiangAMD
opened
2 months ago
0
[Issue]: RuntimeError: FlashAttention forward only supports head dimension at most 128
#51
xxtars
closed
1 month ago
2
[Issue]: Error in the implementation ?
#50
PierreColombo
opened
3 months ago
0
add benchmark script
#49
fsx950223
closed
3 months ago
2
add FA api benchmark csv
#48
fsx950223
opened
3 months ago
1
GPUAI-1250 - Flash Attention v2.04 module rotary cannot be used code fixed
#47
xiaoxiangAMD
opened
4 months ago
2
aac.amd: MI210 - roberta-large with sequence length 8192 and batch_size 1 fails
#46
michaelfeil
opened
4 months ago
0
[Feature]: Is there a Flash-Decoding algorithm implemented based on Composable kernel?
#45
zhangxiao-stack
opened
4 months ago
3
[Issue]: Backward performance
#44
netw0rkf10w
opened
4 months ago
1
[Issue]: Unstable training
#43
netw0rkf10w
opened
4 months ago
1
[Issue]: Installation failed through Dockerfile
#42
amdrenwuli
opened
4 months ago
4
[Issue]: RuntimeError: Expected dout_seq_stride == out_seq_stride to be true, but got false.
#41
donglixp
opened
4 months ago
11
[Issue]: Expected dout_seq_stride == out_seq_stride to be true, but got false
#40
ehartford
opened
5 months ago
13
Installation error
#39
ekazakos
opened
5 months ago
2
Allow gfx908 to build
#38
luizanao
closed
4 months ago
0
Support for MI100 gfx908
#37
luizanao
closed
4 months ago
0
Another installation error
#36
ekazakos
closed
5 months ago
1
Merge to upstream flash-attention repo
#35
ehartford
opened
5 months ago
9
Support for other modules (rotary, xentropy, layer_norm)
#34
bbartoldson
opened
5 months ago
4
replace kernel implementation using CK tile-programming performant kernels
#33
carlushuang
opened
5 months ago
1
Not working on MI250
#32
PierreColombo
closed
5 months ago
0
undefined symbol: hipGetDevicePropertiesR0600
#31
alain40
opened
6 months ago
2
can mask be supported?
#30
unwritten
opened
6 months ago
0
Mi50 Support
#29
YehowshuaScaled
opened
6 months ago
2
installation error
#28
donglixp
opened
6 months ago
14
RDNA3 support
#27
WilliamGazeley
opened
6 months ago
62
Is this v2 or v1?
#26
netw0rkf10w
closed
6 months ago
4
installation error of Method 1 with the recommended docker
#25
donglixp
closed
6 months ago
4
MI100 Support
#24
LoggerHead22
opened
7 months ago
13
Make installation steps look better
#23
Naomiusearch
closed
6 months ago
0
Feature request: Sliding Window Attention
#22
tjtanaa
opened
7 months ago
6
Support mfma_f32_16x16x16f16
#21
hclearner
closed
6 months ago
6
Install failed
#20
1787648106
opened
7 months ago
12
Remove Hardcoded Building Options
#19
howiejayz
closed
7 months ago
2
Remove offload-arch=native in the build
#18
fxmarty
closed
7 months ago
6
Ifu mqa
#17
guangzlu
closed
8 months ago
0
Next