Fix QwenImage txt_seq_lens handling #12702
sayakpaul
merged
59 commits into
huggingface:main
from
kashif:txt_seq_lens
Fix QwenImage txt_seq_lens handling
b547fcf1
formatting
72a80c66
formatting
88cee8b5
dxqb
commented
on 2025-11-29
remove txt_seq_lens and use bool mask
ac5ac24d
Merge branch 'main' into txt_seq_lens
04775260
use compute_text_seq_len_from_mask
18efdded
add seq_lens to dispatch_attention_fn
6a549d45
use joint_seq_lens
2d424e03
remove unused index_block
30b5f988
yiyixuxu
commented
on 2025-12-05
Merge branch 'main' into txt_seq_lens
588dc049
WIP: Remove seq_lens parameter and use mask-based approach
f1c2d996
Merge branch 'txt_seq_lens' of https://github.com/kashif/diffusers in…
ec524175
fix formatting
beeb0206
undo sage changes
5c6f8e39
xformers support
5d434f63
hub fix
71ba6036
Merge branch 'main' into txt_seq_lens
babf490b
fix torch compile issues
afad3357
Merge branch 'main' into txt_seq_lens
2d5ab166
fix tests
c78a1e96
use _prepare_attn_mask_native
d6d4b1d1
proper deprecation notice
e999b769
add deprecate to txt_seq_lens
8115f0b2
yiyixuxu
commented
on 2025-12-09
yiyixuxu
commented
on 2025-12-09
Update src/diffusers/models/transformers/transformer_qwenimage.py
3b1510c3
Update src/diffusers/models/transformers/transformer_qwenimage.py
3676d8e7
Only create the mask if there's actual padding
9ed0ffd5
Merge branch 'main' into txt_seq_lens
abec4615
fix order of docstrings
e26e7b36
yiyixuxu
approved these changes
on 2025-12-10
Adds performance benchmarks and optimization details for QwenImage
59e38829
Merge branch 'main' into txt_seq_lens
0cb2138d
rope_text_seq_len = text_seq_len
60bd4543
rename to max_txt_seq_len
a5abbb8c
Merge branch 'main' into txt_seq_lens
8415c57d
Merge branch 'main' into txt_seq_lens
afff5b7d
Merge branch 'main' into txt_seq_lens
8dc6c3f3
removed deprecated args
22cb03db
undo unrelated change
125a3a41
Updates QwenImage performance documentation
b5b63421
Updates deprecation warnings for txt_seq_lens parameter
61f52658
fix compile
2ef38e2c
Merge branch 'txt_seq_lens' of https://github.com/kashif/diffusers in…
270c63fa
formatting
35efa063
fix compile tests
50c48152
Merge branch 'main' into txt_seq_lens
c88bc063
rename helper
1433783b
remove duplicate
8de799cb
sayakpaul
commented
on 2025-12-18
smaller values
fc937471
Merge branch 'main' into txt_seq_lens
8bb47d8c
removed
b7c288a3
Merge branch 'main' into txt_seq_lens
4700b7fe
dxqb
commented
on 2025-12-21
dxqb
commented
on 2025-12-21
use torch.cond for torch compile
4fe76594
Construct joint attention mask once
77902bcf
test different backends
5b570c79
Merge branch 'main' into txt_seq_lens
4d4e5f48
construct joint attention mask once to avoid reconstructing in every …
8ee4d958
Update src/diffusers/models/attention_dispatch.py
11e0e453
formatting
9d760740
Merge branch 'main' into txt_seq_lens
67a7d242
raising an error from the EditPlus pipeline when batch_size > 1
d20222d9
Login to write a write a comment.
Login via GitHub