transformers

Fix duplicate flash-attention kwargs in Qwen omni/VL encoders

#47823

Closed