COMMITS
September 10, 2026
S
[Performance] Optimize strided transpose and fp8 1x128 quantization (#79743)
SUN Dong committed
O
[Operator Mechanism] Remove unbind int32 element limit (#79737)
omoYang committed
O
[Operator Mechanism ] Add compatible logsumexp cuda kernel (#79732)
omoYang committed
O
O
[Operator Mechanism] Align cosine_similarity with torch2.12 (#79710)
omoYang committed
September 9, 2026
A
[Distributed] Add zero-SM all_gather and all-to-all (#79686)
adam-xiaoyao committed
B
fix test standard (#79708)
Bohao Fan committed
September 8, 2026
Z
[FSDP] optim FSDP save/load (#79724)
ZhenxingLi committed
September 5, 2026
September 3, 2026
O
update approval from luotao1 to wanghuancoder (#79735)
omoYang committed
September 2, 2026
W
[FlexCheckpoint] Add AOA statement generation support (#79723)
wangbingguang committed
P
[Docs] Clarify PR title review guidance (#79733)
Paddle-Bot committed
Z
[FSDP] Fix FSDP perf (#79713)
ZhenxingLi committed
September 1, 2026
L
[FlexCheckpoint] Add LoadTransform extension point for quantized checkpoints (#79712)
Liumengyuan committed
X
add symmetric gemm for muon update (#79694)
Xiangrui Yu committed
August 31, 2026
G
[CI] Update Ubuntu 20 and manylinux CUDA matrix (#79602)
gouzil committed
L
[CI] Allow Paddle-Bot approval in bot gate (#79714)
liuhao2638 committed
G
[Build] Require GCC 11 for C++20 baseline (#79720)
gouzil committed
W
[Environment Adaptation] Upgrade NCCL to 2.31.2 (#79705)
Wang Shigure committed
L
[FlexCheckpoint] Fix nested state_dict load overwritten by loop variable (#79707)
Liumengyuan committed
August 28, 2026
N
[CI] Update approval check list (#79699)
Nyakku Shigure committed
August 27, 2026
H
p2p overlap with dw and recompute (#79681)
Haze188 灏喆 committed
August 26, 2026
G
[Build] Enable CUDA C++20 with host code isolation (#79656)
gouzil committed
F
fix topk (#79693)
feixi committed
Y
Fix truncated_gaussian_random bug in large tensor (#79695)
Yichen Zhang committed
B
align uniform with torch in large tensor (#79665)
Bohao Fan committed
August 25, 2026
S
W
fix muon has bf16 and fp32 params bug (#79687)
Wennie396 committed