DeepSpeed
[Ulysses] Add opt-in uneven-shard rejection and FPDT padding
#9059
Merged

[Ulysses] Add opt-in uneven-shard rejection and FPDT padding #9059

jschneider-canva
jschneider-canva fix(sequence): validate uneven Ulysses shards before attention collec…
d737176b
jschneider-canva fix(sequence): pad FPDT inputs to the global chunk size
5a8b4a67
jschneider-canva jschneider-canva marked this pull request as ready for review 7 days ago
jschneider-canva jschneider-canva requested a review from tohtana tohtana 7 days ago
jschneider-canva jschneider-canva requested a review from tjruwase tjruwase 7 days ago
jschneider-canva jschneider-canva requested a review from loadams loadams 7 days ago
chatgpt-codex-connector
chatgpt-codex-connector
chatgpt-codex-connector commented on 2026-11-12
jschneider-canva docs(sequence): document FPDT padding and loss normalization
a170cbbc
jschneider-canva
tohtana
tohtana requested changes on 2026-11-12
jschneider-canva docs(sequence): demonstrate safe masking of padded FPDT losses
f24d695d
jschneider-canva jschneider-canva requested a review from tohtana tohtana 6 days ago
tohtana tohtana added needs-gpu-test
tohtana
tohtana approved these changes on 2026-11-13
tohtana tohtana enabled auto-merge 5 days ago
tohtana tohtana merged 663de973 into master 5 days ago
tohtana tohtana removed needs-gpu-test

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone