transformers

Continuous batching for linear attention layers

#50350

Open