[GPU][MTL] Resolve long token performance regression in MTL 125H plat… #15573

Job	Run time
Smart_CI	18s
Docker	55s
Build / Build	10m 49s
OpenVINO JS API	0s
Samples	0s
Conformance	0s
ONNX Models Tests	0s
ONNX Runtime Integration	0s
CPU functional tests	0s
iGPU Tests	0s
NVIDIA plugin	0s
OpenVINO tokenizers extension	0s
dGPU Tests	0s
Debian Packages / Debian Packages	1m 41s
C++ unit tests / C++ unit tests	55s
Python API tests / Python API tests	0s
Python unit tests / Python unit tests	2m 44s
Pytorch Layer Tests / PyTorch Layer Tests	3m 13s
TensorFlow Layer Tests	0s
TensorFlow Hugging Face Models tests	0s
TensorFlow TF Hub Models tests	0s
TensorFlow Models tests	0s
PyTorch Models tests	0s
PyTorch Models Nightly Scope1 tests	0s
PyTorch Models Nightly Scope2 tests	0s
JAX/Flax Models tests	0s
ci/gha_overall_status	1s
	20m 36s

Provide feedback