[E2E] Torchbench models got fail_accuracy

[mengfei25]

🐛 Describe the bug

python benchmarks/dynamo/torchbench.py --accuracy --bfloat16 -d xpu -n10 --inference  --only shufflenet_v2_x1_0 --backend=inductor

============ Summary for torchbench bfloat16 inference accuracy ============
Real failed models: 14 [['timm_nfnet', 'fail_accuracy'], ['shufflenet_v2_x1_0', 'fail_accuracy'], ['mobilenet_v2', 'fail_accuracy'], ['timm_resnest', 'fail_accuracy'], ['mnasnet1_0', 'fail_accuracy'], ['timm_vovnet', 'fail_accuracy'], ['resnet152', 'fail_accuracy']]
============ Summary for torchbench bfloat16 training accuracy ============
Real failed models: 11 [['shufflenet_v2_x1_0', 'fail_accuracy'], ['mnasnet1_0', 'fail_accuracy'], ['timm_nfnet', 'fail_accuracy'], ['mobilenet_v2', 'fail_accuracy'], ['resnet152', 'fail_accuracy'], ['timm_resnest', 'fail_accuracy'], ['timm_vovnet', 'fail_accuracy']]
============ Summary for torchbench float16 training accuracy ============
Real failed models: 5 [['timm_regnet', 'fail_accuracy']
============ Summary for torchbench amp_bf16 inference accuracy ============
Real failed models: 17 [['resnet152', 'fail_accuracy'], ['resnet18', 'fail_accuracy'], ['timm_resnest', 'fail_accuracy'], ['mobilenet_v2', 'fail_accuracy'], ['timm_vovnet', 'fail_accuracy'], ['shufflenet_v2_x1_0', 'fail_accuracy'], ['mnasnet1_0', 'fail_accuracy'], ['timm_regnet', 'fail_accuracy'], ['resnext50_32x4d', 'fail_accuracy'], ['densenet121', 'fail_accuracy']]
============ Summary for torchbench amp_bf16 training accuracy ============
Real failed models: 14 [['resnet18', 'fail_accuracy'], ['shufflenet_v2_x1_0', 'fail_accuracy'], ['resnet50', 'fail_accuracy'], ['timm_resnest', 'fail_accuracy'], ['mnasnet1_0', 'fail_accuracy'], ['densenet121', 'fail_accuracy'], ['timm_regnet', 'fail_accuracy'], ['mobilenet_v2', 'fail_accuracy'], ['resnext50_32x4d', 'fail_accuracy'], ['resnet152', 'fail_accuracy'], ['timm_vovnet', 'fail_accuracy']]
============ Summary for torchbench amp_fp16 training accuracy ============
Real failed models: 5 [['timm_regnet', 'fail_accuracy']]

Versions

env:
python: 3.10
XPU_OPS: 9ed0a1a
TRITON_COMMIT_ID: e98b6fcb8df5b44eb0d0addb6767c573d37ba024
TORCH_COMMIT_ID: 4f8b7c4272db521f7ffc4070ce1bdece513d1183
TORCHBENCH_COMMIT_ID: 03cde49eba0580ed17f9ae2250832fd8af4ed756
TORCHVISION_COMMIT_ID: d23a6e1664d20707c11781299611436e1f0c104f
TORCHAUDIO_COMMIT_ID: a6b0a140cc13216975e8922093459019537bb80a
TRANSFORMERS_VERSION: 243e186efbf7fb93328dd6b34927a4e8c8f24395
TIMM_COMMIT_ID: ac3470188b914c5d7a5058a7e28b9eb685a62427
DRIVER_VERSION: 1.23.10.49.231129.50
KERNEL_VERSION: 5.15.0-73-generic #80-Ubuntu SMP Mon May 15 15:18:26 UTC 2023
BUNDLE_VERSION: 2025.0.1.20241113
OS_PRETTY_NAME: Ubuntu 22.04.2 LTS
GCC_VERSION: 11

[mengfei25]

Caused by e035f6b