[`core` ] Integrate Flash attention 2 in most used models #25598
v1
8bb77a1c
oops
2e18421a
working v1
fe5795ec
fixup
4bd15e2d
add some TODOs
49fe3185
Merge branch 'add-flash-attn-2' of https://github.com/younesbelkada/t…
f5d440b1
fixup
50491e83
younesbelkada
changed the title [`PoC` / `core` ] Attempt to integration Flash attn 2 in most used models [`PoC` / `core` ] Attempt to integrate Flash attn 2 in most used models 2 years ago
Merge remote-tracking branch 'upstream/main' into add-flash-attn-2
7df78c09
padding support + try with module replacement
0e30d13c
nit
ad8b9054
alternative design
3c31f10e
oops
2628bf3c
Merge remote-tracking branch 'upstream/main' into HEAD
56d0b49c
add `use_cache` support for llama
20d1b370
v1 falcon
a82f1ca4
nit
c72e8ff8
a bit of refactor
66823f92
nit
41f8f3d5
nits nits
a64a1a93
add v1 padding support falcon (even though it seemed to work before)
67e3fc29
nit
8444ab66
falcon works
8b1c2df2
fixup
c3ebcd2e
v1 tests
1c212d8d
nit
46187011
fix generation llama flash
85ec9467
Merge branch 'add-flash-attn-2' of https://github.com/younesbelkada/t…
2248f20d
update tests
0881ced1
Merge branch 'add-flash-attn-2' of https://github.com/younesbelkada/t…
a8a1b2df
fix tests + nits
2be3e03f
fix copies
b6d3e582
fix nit
b47e85c7
test- padding mask
db8bd640
stype
58848ab3
add more mem efficient support
3f73557d
Update src/transformers/modeling_utils.py
baae7363
fixup
55f61406
Merge remote-tracking branch 'upstream/main' into add-flash-attn-2
10d5c1b7
nit
3fb221a2
fixup
a931aeb9
remove it from config when saving
68a12040
fixup
36e0d6e0
revert docstring
2beeb680
add more checks
7b5da2c5
use values
b99a582e
oops
adaed459
new version
7f06af62
fixup
2d36c6f1
Merge branch 'main' into add-flash-attn-2
a663fa4f
add same trick for falcon
9d3693f8
nit
65ae59c8
add another test
43185b5e
change tests
c61157e7
fix issues with GC and also falcon
2f177925
fixup
65c38617
oops
165a5030
Update src/transformers/models/falcon/modeling_falcon.py
5abc702f
add init_rope
5069e4a5
Merge branch 'add-flash-attn-2' of https://github.com/younesbelkada/t…
11400d89
updates
ace7939a
fix copies
fe9b16d8
Merge branch 'main' into add-flash-attn-2
6174c06f
fixup
acfc954f
fixup
33a0f629
more clarification
ee8ba206
fixup
e28fb0bf
right padding tests
025727c0
add docs
8f7e4008
add FA in docker image
3259392d
more clarifications
57a077b7
add some figures
e62b0b86
add todo
74194383
younesbelkada
marked this pull request as ready for review 2 years ago
rectify comment
3ba5e983
Change to FA2
585e4634
Update docs/source/en/perf_infer_gpu_one.md
ec0f8b9f
split in two lines
3e5ea353
change test name
4bb1bc53
Merge remote-tracking branch 'origin/main' into add-flash-attn-2
3ea46338
add more tests
b67c21e6
some clean up
5b735574
remove `rearrange` deps
48e3bcff
add more docs
04613840
younesbelkada
changed the title [`PoC` / `core` ] Attempt to integrate Flash attn 2 in most used models [`core` ] Attempt to integrate Flash attn 2 in most used models 2 years ago
revert changes on dockerfile
8d72a66b
Revert "revert changes on dockerfile"
73b2f073
revert changes on dockerfile
fb7654cf
Apply suggestions from code review
a737bdea
address some comments
80951ae1
docs
6f7ff42b
use inheritance
257a6335
Update src/transformers/testing_utils.py
360da707
fixup
1d91bc4a
Merge branch 'main' into add-flash-attn-2
8ecab971
Apply suggestions from code review
7c5720f1
Update src/transformers/modeling_utils.py
28b82e20
final comments
84b57938
clean up
949172f1
style
825c7e0f
add cast + warning for PEFT models
1af232cf
fixup
d7f16c58
younesbelkada
changed the title [`core` ] Attempt to integrate Flash attn 2 in most used models [`core` ] Integrate Flash attn 2 in most used models 2 years ago
younesbelkada
changed the title [`core` ] Integrate Flash attn 2 in most used models [`core` ] Integrate Flash attention 2 in most used models 2 years ago
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub