[Dev] Support paged stash with device-initiated GroupedTensor experts - #6845

Draft

lhb8125 wants to merge 10 commits into

lhb8125:denliu/paged-stash-grouped-tensor-dev

Commits on Aug 25, 2026

[Main] Support device-init grouped linear module too with when not using TE opfuser (#6000)

zhongbozhu
authored and
lhb8125
committed
1a16d50
feat: support paged stash with GroupedTensor experts
lhb8125
committed
4347f36
refactor: use TE paged stash markers
lhb8125
committed
d631ce6
refactor: require fused activations for paged stash
lhb8125
committed
5c7e479
refactor: minimize paged stash integration diff
lhb8125
committed
2b2c261
refactor: narrow paged stash marker propagation
lhb8125
committed
e2dbbd7
refactor: route paged stash markers through TE adapter
lhb8125
committed
a322707
docs: trim paged stash implementation details
lhb8125
committed
34d3765
chore: align dev copyright headers
lhb8125
committed
f38e49f
test: gate grouped-tensor paged stash on TE support
lhb8125
committed
e569fa3