[DiffusionGemma] Support gradient checkpointing #46572
zucchini-nlp
merged
16 commits into
huggingface:main
from
kashif:diffusion-gemma-gradient-checkpointing
kashif
changed the title
kashif
changed the title 
zucchini-nlp
commented
on 2026-06-12
ArthurZucker
commented
on 2026-06-12
ArthurZucker
commented
on 2026-06-17
Support gradient checkpointing in DiffusionGemma
00ee1d04
Test that checkpointing reproduces the exact loss and gradients
e039af9a
Merge branch 'main' into diffusion-gemma-gradient-checkpointing
f987976a
zucchini-nlp
commented
on 2026-06-24
Merge branch 'main' into diffusion-gemma-gradient-checkpointing
35d79b30
update use_cache test now that the decoder ignores it instead of raising
92555cd6
Merge branch 'main' into diffusion-gemma-gradient-checkpointing
a371e48e
Cyrilvallez
commented
on 2026-07-08
Merge remote-tracking branch 'upstream/main' into diffusion-gemma-gra…
92ee0bd9
use a single read-only kv cache under gradient checkpointing
1332f74e
Merge remote-tracking branch 'upstream/main' into diffusion-gemma-gra…
0ba30cdb
Merge remote-tracking branch 'kashif/diffusion-gemma-gradient-checkpo…
beee2ceb
drop the unused encoder kv return and guard the decoder cache
edf9a694
Merge branch 'main' into diffusion-gemma-gradient-checkpointing
660a958d
Merge branch 'main' into diffusion-gemma-gradient-checkpointing
414274c4
reuse BaseModelOutputWithPast and drop the unrelated mask cast
e7abc989
Merge branch 'main' into diffusion-gemma-gradient-checkpointing
6858e113
Merge branch 'main' into diffusion-gemma-gradient-checkpointing
2812f155
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub