Add Granite-swa and Granitemoe-swa model support #47179
vasqu
merged
34 commits into
huggingface:main
from
daviswer:granite-swa
Add GraniteSWA model: Granite with Sliding Window Attention + learnab…
3c2e456c
Add FA3 attention dispatch with native LSE for sink scaling
8fdf0adb
Enable DynamicSlidingWindowCache for memory-efficient KV cache
1fe06100
Finalize GraniteSWA: modular impl, FA4 support, formatting, tests, docs
895dc30e
Add GraniteMoeSWA (MoeShared+SWA)
eec43fba
Update model refs, add per-layer rope/nope support a la Llama4
56a2274c
Small formatting/date fixes
c40cf934
Test comments
ddada049
Merge branch 'main' into granite-swa
76d88f7f
automodel import style fix
da94124a
Fix automodel import order after pull
f77b683e
Regenerate modeling_ files post-resync to main
e9baf4d6
Fix upload date in model cards
047cd7ef
Additional TP / historic test exclusions post-sync to main
6a6f0533
Docs cleanup, one final round on submit date
4e99d5ac
Fix grante_swa init and imports
e0d10d43
Addressing review comments:
e04cd0bb
Replace binary rope/nope support with dynamic rope freq support
91e66639
Reorder router outputs to support native EP (and tests)
c984f0b0
Clean flex flags, flash correction tests, rope constructors
86b57162
Correct Flash2 test skip check
5f04c023
Clean docstrings/comments, fix flash2 causal lm test
1b27b6be
Merge branch 'main' into granite-swa
b6c0c828
Modelcard dates
28604aa3
vasqu
approved these changes
on 2026-07-23
Fix rope typing and slow test shape checking
22c18490
Merge branch 'main' into granite-swa
fca719cc
zucchini-nlp
commented
on 2026-07-24
Update integration test expectations by device
e8c6337e
Merge branch 'main' into granite-swa
abfb6565
Regenerate modeling files to match inherited modular upstream changes
a866b4ab
adjust to ci
5474d5bc
fixup repo
2cee0aec
Merge branch 'main' into granite-swa
8b242a63
try
d4cf007c
kernels sync
c3ba1d45
Login to write a write a comment.
Login via GitHub