[E2E] Timm models got fail_accuracy

[mengfei25]

🐛 Describe the bug

Failed dtype: bfloat16 and amp_bf16. float32, float16 and amp_fp16 passed
python benchmarks/dynamo/timm_models.py --accuracy --float16 -d xpu -n10 --training--only tf_efficientnet_b0 --backend=inductor
============ Summary for timm_models bfloat16 inference accuracy ============
Real failed models: 13 [['tf_efficientnet_b0', 'fail_accuracy'], ['spnasnet_100', 'fail_accuracy'], ['inception_v3', 'fail_accuracy'], ['regnety_002', 'fail_accuracy'], ['dla102', 'fail_accuracy'], ['dpn107', 'fail_accuracy'], ['hrnet_w18', 'fail_accuracy'], ['lcnet_050', 'fail_accuracy'], ['swsl_resnext101_32x16d', 'fail_accuracy'], ['fbnetc_100', 'fail_accuracy'], ['ghostnet_100', 'fail_accuracy'], ['dm_nfnet_f0', 'fail_accuracy'], ['mnasnet_100', 'fail_accuracy']]
============ Summary for timm_models bfloat16 training accuracy ============
Real failed models: 30 [['lcnet_050', 'fail_accuracy'], ['mixnet_l', 'fail_accuracy'], ['regnety_002', 'fail_accuracy'], ['res2next50', 'fail_accuracy'], ['ese_vovnet19b_dw', 'fail_accuracy'], ['convmixer_768_32', 'fail_accuracy'], ['mobilenetv3_large_100', 'fail_accuracy'], ['fbnetv3_b', 'fail_accuracy'], ['repvgg_a2', 'fail_accuracy'], ['selecsls42b', 'fail_accuracy'], ['swsl_resnext101_32x16d', 'fail_accuracy'], ['tf_efficientnet_b0', 'fail_accuracy'], ['dla102', 'fail_accuracy'], ['dm_nfnet_f0', 'fail_accuracy'], ['mnasnet_100', 'fail_accuracy'], ['inception_v3', 'fail_accuracy'], ['mobilenetv2_100', 'fail_accuracy'], ['spnasnet_100', 'fail_accuracy'], ['visformer_small', 'fail_accuracy'], ['adv_inception_v3', 'fail_accuracy'], ['gluon_inception_v3', 'fail_accuracy'], ['ghostnet_100', 'fail_accuracy'], ['res2net50_14w_8s', 'fail_accuracy'], ['hrnet_w18', 'fail_accuracy'], ['pnasnet5large', 'fail_accuracy'], ['convnext_base', 'fail_accuracy'], ['tf_mixnet_l', 'fail_accuracy'], ['rexnet_100', 'fail_accuracy'], ['res2net101_26w_4s', 'fail_accuracy'], ['fbnetc_100', 'fail_accuracy']]
============ Summary for timm_models amp_bf16 inference accuracy ============
Real failed models: 19 [['hrnet_w18', 'fail_accuracy'], ['swsl_resnext101_32x16d', 'fail_accuracy'], ['regnety_002', 'fail_accuracy'], ['lcnet_050', 'fail_accuracy'], ['rexnet_100', 'fail_accuracy'], ['gluon_inception_v3', 'fail_accuracy'], ['adv_inception_v3', 'fail_accuracy'], ['dla102', 'fail_accuracy'], ['ghostnet_100', 'fail_accuracy'], ['mnasnet_100', 'fail_accuracy'], ['mobilenetv2_100', 'fail_accuracy'], ['fbnetc_100', 'fail_accuracy'], ['res2net101_26w_4s', 'fail_accuracy'], ['res2net50_14w_8s', 'fail_accuracy'], ['dpn107', 'fail_accuracy'], ['inception_v3', 'fail_accuracy'], ['res2next50', 'fail_accuracy'], ['spnasnet_100', 'fail_accuracy'], ['tf_efficientnet_b0', 'fail_accuracy']]
============ Summary for timm_models amp_bf16 training accuracy ============
Real failed models: 29 [['convmixer_768_32', 'fail_accuracy'], ['fbnetc_100', 'fail_accuracy'], ['hrnet_w18', 'fail_accuracy'], ['ese_vovnet19b_dw', 'fail_accuracy'], ['ghostnet_100', 'fail_accuracy'], ['regnety_002', 'fail_accuracy'], ['tf_mixnet_l', 'fail_accuracy'], ['mixnet_l', 'fail_accuracy'], ['repvgg_a2', 'fail_accuracy'], ['selecsls42b', 'fail_accuracy'], ['rexnet_100', 'fail_accuracy'], ['res2net101_26w_4s', 'fail_accuracy'], ['res2next50', 'fail_accuracy'], ['fbnetv3_b', 'fail_accuracy'], ['swsl_resnext101_32x16d', 'fail_accuracy'], ['adv_inception_v3', 'fail_accuracy'], ['spnasnet_100', 'fail_accuracy'], ['mobilenetv3_large_100', 'fail_accuracy'], ['inception_v3', 'fail_accuracy'], ['visformer_small', 'fail_accuracy'], ['gluon_inception_v3', 'fail_accuracy'], ['lcnet_050', 'fail_accuracy'], ['tf_efficientnet_b0', 'fail_accuracy'], ['dla102', 'fail_accuracy'], ['mobilenetv2_100', 'fail_accuracy'], ['mnasnet_100', 'fail_accuracy'], ['pnasnet5large', 'fail_accuracy'], ['eca_halonext26ts', 'fail_accuracy'], ['res2net50_14w_8s', '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

[mengfei25]

Most are fixed, and only the following are still failed
============ Summary for timm_models bfloat16 training accuracy ============
Real failed models: 1 [['convnext_base', 'fail_accuracy']]
============ Summary for timm_models amp_bf16 training accuracy ============
Real failed models: 2 [['fbnetv3_b', 'fail_accuracy'], ['eca_halonext26ts', 'fail_accuracy']]