transformers
🚨 [`Kernels`] Refactor all linear attn models & native kernels fallback
#47630
Merged

🚨 [`Kernels`] Refactor all linear attn models & native kernels fallback #47630

vasqu merged 48 commits into huggingface:main from vasqu:kernels-fallback
vasqu
Cyrilvallez kernel native
f438a2e8
Cyrilvallez fix warning
7c1335b8
Cyrilvallez have to fix other modular to be coherent
f8ff0db0
Cyrilvallez remove useless
9a8127e8
Cyrilvallez use native lib as well
72b45caf
Cyrilvallez modular
54d80dfb
Cyrilvallez combine them
d2beaf64
Cyrilvallez doc
c256c47b
vasqu ignore kwargs, allow og
30e0ff74
vasqu Merge remote-tracking branch 'upstream/main' into kernels-fallback
c57b520a
vasqu fix
38c2ece9
vasqu gdn like paths (missing conv of olmo hybrid)
7788cac4
HuggingFaceDocBuilderDev
vasqu olmo hybrid fused conv style
cf49de64
vasqu fix
73db89ba
vasqu fix
88966967
vasqu poc mamba2, kernel must compile but torch seems to match
d3b27a66
vasqu kernels match --> hf kernels will be needed
f0aa5812
vasqu style
263f3077
vasqu quick fixes
a189da9e
vasqu Merge remote-tracking branch 'upstream/main' into kernels-fallback
8134d382
vasqu mamba2 works
4ebeb111
vasqu mamba2 suite of models
1786c75b
vasqu oops
0eca0d07
vasqu conv1ds across other models
82bfadec
vasqu let's try this
520e340b
vasqu mamba base implementation
f69acfed
Cyrilvallez
Cyrilvallez commented on 2026-07-31
Cyrilvallez
Cyrilvallez commented on 2026-07-31
vasqu fixups as per review comments
bfbf9f15
vasqu fixup mamba tests and other issues
511d4a61
vasqu make no shape check by default (single padded sample should also work)
3ee52c35
vasqu remove todo
34051c4d
vasqu propogate mamba1
c3a6c982
vasqu style
fcf7292f
vasqu fix mambapy + enable on jamba
3823bcbd
vasqu zamba1
380e5fc7
vasqu style
ad78e096
vasqu fix early cast (leads to non fp32 norm)
3c51e11d
vasqu fix padding free path
b05fb522
vasqu
github-actions
vasqu fix offload
98f62d41
vasqu Merge remote-tracking branch 'upstream/main' into kernels-fallback
66cff331
vasqu
github-actions
github-actions
vasqu vasqu marked this pull request as ready for review 18 days ago
github-actions
vasqu Merge remote-tracking branch 'upstream/main' into kernels-fallback
2ec940b7
vasqu
github-actions
vasqu bump kernels
9b266aa3
github-actions
vasqu remove todo
c4b785c8
vasqu avoid onnx export and fix mamba2 test
c85e6220
vasqu
github-actions
vasqu fix bamba test
2b4fd754
github-actions
vasqu update falcon mamba - aligned with all other devices
2e485f3f
vasqu
github-actions
github-actions
vasqu oops
d81829f6
vasqu
vasqu
vasqu commented on 2026-08-04
github-actions
vasqu Merge branch 'main' into kernels-fallback
9dabdf29
github-actions
vasqu
github-actions
github-actions
Cyrilvallez
Cyrilvallez commented on 2026-08-05
Cyrilvallez Cyrilvallez changed the title [`Kernels`] Native fallback 🚨 [`Kernels`] Refactor all mamba's recurrent part & Native kernels fallback 17 days ago
Cyrilvallez Cyrilvallez changed the title 🚨 [`Kernels`] Refactor all mamba's recurrent part & Native kernels fallback 🚨 [`Kernels`] Refactor all mamba models recurrent part & Native kernels fallback 17 days ago
vasqu adress review
4c1864de
github-actions
vasqu vasqu changed the title 🚨 [`Kernels`] Refactor all mamba models recurrent part & Native kernels fallback 🚨 [`Kernels`] Refactor all linear attn models & native kernels fallback 17 days ago
github-actions
Cyrilvallez
Cyrilvallez approved these changes on 2026-08-05
vasqu vasqu merged e91f7ef4 into main 17 days ago
vasqu vasqu deleted the kernels-fallback branch 17 days ago

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
Milestone