issues
search
philipturner
/
metal-flash-attention
FlashAttention (Metal Port)
MIT License
347
stars
18
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Documentation
#23
philipturner
closed
2 weeks ago
0
Mixed precision
#22
philipturner
closed
2 weeks ago
0
Fixed the register pressure issues on M1, and improved performance on M3
#21
philipturner
closed
3 weeks ago
0
C++ reference implementation of GEMM
#20
philipturner
closed
1 month ago
0
Replace the main branch with the rewritten implementation
#19
philipturner
closed
2 months ago
0
Compute Command Encoder Question
#18
jafioti
closed
1 week ago
1
`bfloat16` support
#17
cloneable
closed
1 week ago
5
Feature/larger batches
#16
FL33TW00D
closed
7 months ago
4
Fix grouped query support
#15
liuliu
closed
3 weeks ago
2
simdgroup_async issues - Xcode Version 15.0.1 (15A507) / M3 Max 14.1.2 (23B2091)
#14
bpkeene
closed
8 months ago
2
Allow MFA to be configured to use float32 as accumulator
#13
liuliu
closed
9 months ago
1
Weird performance when using shared memory in GEMV
#12
FdyCN
closed
1 week ago
8
M3 Performance
#11
Narsil
closed
1 week ago
15
Is it possible that header <metal_simdgroup_future> included using JIT compilation?
#10
FdyCN
closed
9 months ago
1
[Question] Why use index 50000 instead of 101?
#9
FdyCN
closed
9 months ago
3
Accuracy issues due to attention_matrix accumulated at half-precision & softmax_scale (alpha) applied after qk
#8
liuliu
closed
1 week ago
1
Bfloat gemm
#7
ivarflakstad
closed
6 months ago
1
Purge and autorelease of MTLBuffers so memory is released more efficiently
#6
ivarflakstad
closed
10 months ago
0
Use simd4 memorylayout when calculating buffer length
#5
ivarflakstad
closed
10 months ago
5
Undefined symbols error
#4
jafioti
closed
1 week ago
1
Guidelines for modifying H3 with metal-flash-attention
#3
okpatil4u
closed
1 week ago
5
how to use this flash-attention in python code ?
#2
Johnson-yue
closed
1 week ago
10
ETA on Dense FlashAttention ?
#1
ghost
closed
1 year ago
5