Support nonpad_kv_seqlen and external kv cache in-place update in Attention-24 (CUDA)#27486

Closed

titaiwangms wants to merge 28 commits intomainfrom

titaiwang/support_nonpad_kv_seqlen

Commits on Feb 27, 2026

Commits on Feb 28, 2026

Add causal mode test variants for TensorScatter + nonpad_kv_seqlen

titaiwangms
and
Copilot
committed
Address P2 review findings: tolerance, GQA TODO, and parameter documentation

titaiwangms
and
Copilot
committed
Add batch=1 edge case test for TensorScatter attention

titaiwangms
and
Copilot
committed

Commits on Mar 2, 2026

Commits on Mar 3, 2026