onnxruntime

[CUDA] Benchmark GQA on popular LLM models

#20646

Merged