onnxruntime

Fix GPT-2 no-past attention fusion for transformers >= 4.27

#27449

Merged