[None][refactor] Flatten thop.attention sequence kwargs + rename rotary_embedding_* to rope_* - #14569

Merged

yuxianq merged 6 commits into

NVIDIA:mainfrom

yuxianq:flatten-attn-args

May 29, 2026

Commits on May 28, 2026

[None][refactor] Flatten 5 sequence kwargs in thop.attention C++ signature
yuxianq
committed
[None][refactor] Rename rotary_embedding_* -> rope_* in thop.attention C++ surface
yuxianq
committed
[None][refactor] Update nanobind binding for flat + renamed thop.attention kwargs
yuxianq
committed
[None][refactor] Flatten + rename kwargs in PyTorch backend thop.attention call
yuxianq
committed
[None][refactor] Update AutoDeploy thop.attention call sites for flat + renamed kwargs
yuxianq
committed
[None][test] Forbid Sequence-typed kwargs at thop.attention boundary
yuxianq
committed