[WS1][kernels] Batch-invariant attention (standard softmax--cuda version) - #240

Merged

Flink-ddd merged 6 commits into

RL-Align:mainfrom

maxiaosong1124:feat/issue-147-cuda-standard-attention

Jul 28, 2026

Commits on Jul 14, 2026

feat(attention): CUDA deterministic standard-softmax attention (#147)
maxiaosong1124
committed
e3365b1

Commits on Jul 21, 2026

Commits on Jul 27, 2026

Merge branch 'main' into feat/issue-147-cuda-standard-attention
maxiaosong1124
authored
1d2acd0
fix(gtest): close attention operator spec
maxiaosong1124
committed
fb7d1d5
fix(attention): address review feedback
maxiaosong1124
committed
040acd9