transformers
[`Attn Masks`] Non-vmap default for attention masks
#41852
Merged

[`Attn Masks`] Non-vmap default for attention masks #41852

vasqu merged 17 commits into huggingface:main from vasqu:non-vmap-masks
vasqu
vasqu atmpt 1
dfcb545d
vasqu fixup masking to work correctly with old torch
b87a139f
HuggingFaceDocBuilderDev
vasqu Merge branch 'main' into non-vmap-masks
9aed30d1
vasqu few changes to make things a bit more cleaner
969dab55
vasqu oopsie
513c8ef7
vasqu fix integer overflow on bidirectional masks via indexing fn
466acaba
vasqu rm executorch workarounds --> still need to handle on sliding etc fns…
bbaf41d8
vasqu typo
65357d9b
vasqu docs, fix older torch inplace issue, proper kwarg handling
aaaaec2b
vasqu chunked works with non vmap and older torch, add warning on non guara…
539bafad
vasqu lift unnecessary restriction on older torch
01848e3b
vasqu vasqu changed the title [`WIP`][`Masking`] Non-vmap default for attention masks [`Attn Masks`] Non-vmap default for attention masks 95 days ago
vasqu Merge branch 'main' into non-vmap-masks
9dc62965
vasqu vasqu marked this pull request as ready for review 95 days ago
vasqu vasqu requested a review from ArthurZucker ArthurZucker 95 days ago
vasqu vasqu requested a review from Cyrilvallez Cyrilvallez 95 days ago
vasqu
vasqu commented on 2025-10-29
vasqu
vasqu commented on 2025-10-29
vasqu
vasqu commented on 2025-10-29
vasqu
vasqu commented on 2025-10-29
vasqu simplify a few things, restrict torch < 2.6 to non-vmap (for now)
17c7a486
vasqu try fix
4e6e799b
vasqu remove unnecessary slicing logic
26b266c4
ArthurZucker
ArthurZucker approved these changes on 2025-11-03
jiqing-feng
Cyrilvallez remove legacy func
1fb7510e
Cyrilvallez
Cyrilvallez approved these changes on 2025-11-10
Cyrilvallez harmonize slightly
4f62c81c
IlyasMoutawwakil
vasqu
vasqu vasqu merged 03538a80 into main 83 days ago
vasqu vasqu deleted the non-vmap-masks branch 83 days ago
vasqu

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone