`bridge.models.stepfun.modelling_step37.attention`#

Step3.7 text-side attention re-exports.

Step3.7’s text decoder reuses Step-3.5’s hybrid full/sliding attention layer (Step35DecoderLayer) verbatim, so this module is a structural alias of qwen_vl/modelling_qwen3_vl/attention.py — it surfaces the Step-3.5 decoder layer under the modelling_step37 namespace without redefining any attention math.