onnxruntime

Add --use_multi_head_attention in transformers fusion

#14198

Merged