transformers
Make Transformers more torch-exportable and dynamo-friendly
#42317
Merged
Merged
Make Transformers more torch-exportable and dynamo-friendly #42317
ArthurZucker
merged
70 commits into
main
from
export-friendly
make vlms export friendly
1ab6216a
seq2seq lms
25456f08
biogpt
6d4f744f
more vlms
b8135b4e
colqwen2
2d3f986a
vision models
7e03b171
more vlms
a58b0494
more vlms
e776c188
more vlms
5377e2c5
vectorized vision embedding
6b4c37da
fixup
44e8e47d
more vlms
df1559f2
more vlms
bb6b8a38
generate_masks_with_special_tokens_and_transfer_map
b606772d
custom torch_check
fdc9535f
use custom torch_check
0184d208
revert grounding dino changes
a4d754ec
fixup
75c0fd35
remove file
713c8956
undo
2a4081bb
undo
73cd1baa
IlyasMoutawwakil
commented
on 2025-11-21
IlyasMoutawwakil
commented
on 2025-11-21
copilot-pull-request-reviewer
commented
on 2025-11-21
testing
8aea2bb9
fixes
d3213195
Merge branch 'main' into export-friendly
dd5dc7a0
standard error message
52515c3d
use torch._check_with to raise value error instead of torch._check's …
7cbf4c4f
fix recurrent gemma
bf2578d7
only itemize tensors
f33368da
use spatial shapes list instead of tensor
4ad0bace
fix udop use_cache default value
0241d53d
use tracable condition for seq2seq lms
e55f3836
IlyasMoutawwakil
commented
on 2025-11-24
echarlaix
approved these changes
on 2025-11-27
Merge branch 'main' into export-friendly
2eb2ed17
Merge branch 'main' into export-friendly
942dadf5
make smolvlm exportable
45f88acf
Merge branch 'main' into export-friendly
de845895
fix fastvlm and t5gemma2
25b804fe
Merge branch 'main' into export-friendly
b2ca9e31
fix qwen2_audio and idefics
beeb6729
remove script
3911d411
tbc
76972108
Merge branch 'main' into export-friendly
14113861
skip mra model
8d4d9550
helper
0fcf3e43
Merge branch 'main' into export-friendly
8cf3d2ff
style and document
9cb62d7a
Merge branch 'main' into export-friendly
ada12be8
fix
10bd5198
set experts impl to batched
e8abb270
make xmod exportable and efficient
f4431d66
make more ssms exportable
e04a4746
fix
44316335
revert recurrent gemma
faa4540b
skip models that use chunked attention or rope_index
01d93a20
qwen3_next
23f4485d
assert async
60050d89
Merge branch 'main' into export-friendly
526fb5c8
tensorize (mm) grounding dino mask generation
799ad58a
IlyasMoutawwakil
changed the title
Merge branch 'main' into export-friendly
c551dd01
Merge branch 'main' into export-friendly
ae787661
style
a6d56f1f
fix repo
04dee522
address comments
d3e145da
Merge branch 'main' into export-friendly
ac7c941d
fix qwen2 audio and vits checks
86df05d0
skip two models using kernels by default
e98872d2
skip granite moe hybrid using custom kernels
b9f20c26
disable mamba kernels
d20b998c
vits splinter and videomae
b2884efe
Merge branch 'main' into export-friendly
d45e2cd2
Merge branch 'main' into export-friendly
99be85b4
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub