[CUDA] Add option to use DecoderMaskedMultiheadAttention in BeamSearch #14990
Initial commit
4d41bf5b
Change - 1
d56afff8
Change - 2
80bb3c4b
Change - 3
efab199f
Change -4
058c5bb5
Change - 5
89b7c2bb
Change - 6
f915a393
Change - 7
2ebdd2b2
Change - 8
946371ae
Change - 9
9932efc5
Change - 9.1
1e9000fb
Change - 9.2
b462a647
Change - 10
4419828a
Change - 11
1ebd0e30
Change - 12
6e6a7408
Change - 13
701b4be1
Change - 14
1c612310
Change - 15
4e21961f
Change - 16
ad57765c
Change - 17
95629f0e
More changes
3bfac9bf
Fix builds
acd6e4ce
Fix builds
09c123aa
Support head size 128
f2a013ae
Nit
96504bfb
Try fix builds
46844e2b
Add initial support for BeamSearch
058c3fef
Beam Search commit 2
b7df0556
Beam Search commit 2.1
72ec735f
Beam Search commit 3
66e90897
Beam Search commit 4
7dbca89b
More changes
9f36c871
Nits
37bb8fe2
Fix build
e8238520
Fix builds
99b3729d
Update docs
1e38cc29
yufenglee
dismissed these changes
on 2023-03-14
PR feedback
197d4ba9
hariharans29
dismissed their stale review
via 197d4ba9
3 years ago
wangyems
dismissed these changes
on 2023-03-15
Fix typo
77269108
hariharans29
dismissed their stale review
via 77269108
3 years ago
Fix missing comma
f848125d
Merge remote-tracking branch 'origin/main' into hari/scratch_5
b7d98395
Update docs
8c1bee82
tianleiwu
approved these changes
on 2023-03-15
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub