Commits

History for

Commits on Sep 28, 2026

[FlexCheckpoint] Avoid the full metadata load in the local-resume check (#79807)

DanielSun11
and
claude
authored
e9a8a18

Commits on Sep 22, 2026

[compat] Resolve a scoped spec once instead of re-entering the finder (#79798)
LiYuRio
authored
7c722e5
[Operator Mechanism]Add CUDA small-stride index gradient kernel (#79716)

feixi139
and
claude
authored
8b0a952
[FlexCheckpoint] Simplify AOA name mapping and trim the generation surface (#79791)
wangbingguang2026
authored
6d1e60b

Commits on Sep 21, 2026

Commits on Sep 20, 2026

Commits on Sep 17, 2026

Add machine-balanced partitioning for Muon 2D sharding (#79763)

cszdrg
and
xiaoguoguo626807
authored
01d68e3

Commits on Sep 15, 2026

Commits on Sep 11, 2026

Align CPU var and std with PyTorch (#79747)
feixi139
authored
9a64753

Commits on Sep 10, 2026

[Operator Mechanism] Align cosine_similarity with torch2.12 (#79710)
omoYang
authored
f4e6f68

Commits on Sep 9, 2026

[Distributed] Add zero-SM all_gather and all-to-all (#79686)
adam-xiaoyao
authored
4ac7a75

Commits on Sep 8, 2026

[FSDP] optim FSDP save/load (#79724)
Xing-lil
authored
0ce6198

Commits on Sep 5, 2026

[Performance] Partition Muon 2D params per dtype to restore comm fusion (#79740)
AlAuAu
authored
df1fbe1

Commits on Sep 3, 2026

[Operator Mechanism] Align put_along_axis shape check with torch (#79717)
feixi139
authored
469b33d

Commits on Sep 2, 2026

[FlexCheckpoint] Add AOA statement generation support (#79723)
wangbingguang2026
authored
519f66e
[FSDP] Fix FSDP perf (#79713)
Xing-lil
authored
95aea6a

Commits on Sep 1, 2026

Commits on Aug 31, 2026

Commits on Aug 27, 2026

p2p overlap with dw and recompute (#79681)
Hz188
authored
0489926

Commits on Aug 25, 2026

Commits on Aug 24, 2026

Commits on Aug 20, 2026

Commits on Aug 19, 2026

[Performance] paddle.save supports protocol=5 with zero-copy (#79660)
lshpku
authored
2fd3134
[Build] Support CUDA 13.2 on Windows (#79603)

youge325
and
SigureMo
authored
17780a6

Commits on Aug 17, 2026

[Operator Mechanism] Make view dtype forward only (#79654)
feixi139
authored
5dd4894

Commits on Aug 14, 2026

[Operator Mechanism] Support baddbmm out_dtype for CUDA (#79638)
feixi139
authored
16037ff