transformers
Loading: convert transposed and RoPE-permuted weights before taking the DTensor shard
#48519
Open

Loading: convert transposed and RoPE-permuted weights before taking the DTensor shard #48519

qgallouedec wants to merge 15 commits into main from load-convert-before-shard
qgallouedec
qgallouedec Load: convert transposed and RoPE-permuted weights in full before tak…
bb3a775d
qgallouedec Let conversion ops declare the dims they move; defer the shard only w…
9e2bca95
qgallouedec Keep the decision in one helper
d83a1f72
HuggingFaceDocBuilderDev
vasqu
vasqu
vasqu commented on 2026-09-04
qgallouedec Merge branch 'main' into load-convert-before-shard
324b3116
VI-Arthur
VI-Arthur commented on 2026-09-08
qgallouedec
VI-Arthur
qgallouedec Narrow to the RoPE permutation, leave transposes to shard-on-read
f4562cf1
qgallouedec
qgallouedec Merge branch 'main' into load-convert-before-shard
d8c8613a
qgallouedec Merge branch 'main' into load-convert-before-shard
7d960108
qgallouedec Merge branch 'main' into load-convert-before-shard
717a3a95
VI-Arthur
qgallouedec Merge branch 'main' into load-convert-before-shard
5892c331
qgallouedec Only build the sharding op when it is used
cb61b1aa
qgallouedec Let each op say whether it breaks shard-on-read
071db775
qgallouedec Merge branch 'main' into load-convert-before-shard
5fa9a03d
qgallouedec Merge branch 'main' into load-convert-before-shard
10f60888
qgallouedec Merge branch 'main' into load-convert-before-shard
14fe0333
qgallouedec Merge branch 'main' into load-convert-before-shard
65b6e406
github-actions

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
Milestone