Module Group Offloading #10503
update
d1737e3d
fix
27836699
non_blocking; handle parameters and buffers
6a9a3e59
update
c426a343
DN6
commented
on 2025-01-10
Group offloading with cuda stream prefetching (#10516)
d579037f
Merge branch 'main' into groupwise-offloading
5f336210
update
a8eabd07
Merge branch 'main' into groupwise-offloading
deda9a38
copy model hook implementation from pab
80ac5a72
update; ~very workaround based implementation but it seems to work as…
d2a2981a
more workarounds to make it actually work
01c7d220
cleanup
22aff343
rewrite
42bc19b6
update
8c63bf5a
make sure to sync current stream before overwriting with pinned params
e09e7162
Merge branch 'main' into groupwise-offloading
bf379c1d
better check
0bf0bafc
update
b850c759
remove hook implementation to not deal with merge conflict
6ed9c2f1
Merge branch 'main' into groupwise-offloading
13dd3377
re-add hook changes
073d4bc1
why use more memory when less memory do trick
8ba2bda2
why still use slightly more memory when less memory do trick
b2e838f5
Merge branch 'main' into groupwise-offloading
f30c55fe
optimise
5ea3d8af
add model tests
db2fd3ba
add pipeline tests
a0160e11
update docs
aaa9a534
Merge branch 'main' into groupwise-offloading
17b27533
add layernorm and groupnorm
edf81035
Merge branch 'main' into groupwise-offloading
af62c937
DN6
commented
on 2025-01-31
DN6
added close-to-merge
Merge branch 'main' into groupwise-offloading
f227e156
address review comments
24f92739
improve tests; add docs
8f10d05e
improve docs
06b411fc
Merge branch 'main' into groupwise-offloading
8bd7e3b5
stevhliu
approved these changes
on 2025-02-04
Apply suggestions from code review
904e470f
apply suggestions from code review
3172ed5d
Merge branch 'main' into groupwise-offloading
72aa57f7
update tests
aee24bcc
Merge branch 'main' into groupwise-offloading
db125cea
apply suggestions from review
3f20e6bc
DN6
approved these changes
on 2025-02-06
enable_group_offloading -> enable_group_offload for naming consistency
840576ac
raise errors if multiple offloading strategies used; add relevant tests
8804d746
DN6
commented
on 2025-02-06
handle .to() when group offload applied
954bb7d0
Merge branch 'main' into groupwise-offloading
ba6c4a81
refactor some repeated code
da88c333
remove unintentional change from merge conflict
a872e84e
handle .cuda()
6be43b8a
Merge branch 'main' into groupwise-offloading
274b84e9
DN6
approved these changes
on 2025-02-14
DN6
merged
9a147b82
into main 297 days ago
a-r-r-o-w
deleted the groupwise-offloading branch 297 days ago
Assignees
No one assigned
Labels
close-to-merge
roadmap
Login to write a write a comment.
Login via GitHub