executeV2: Error Code 1: Cask (Cask Pooling Runner Execute Failure)

when I use the lateset expressing for loading the resnet-50, it can run successfully

# 加载预训练的 ResNet-50 模型
model= models.resnet50(ResNet50_Weights.IMAGENET1K_V1)
.......
context.execute_v2(bindings)
cuda.memcpy_dtoh(output_data,d_output)

output:

[[-1.99739981e+00 -8.73850882e-01 -2.61901110e-01 -5.06821036e-01
  -8.42439711e-01 -1.29197025e+00 -1.12298012e+00 -7.87284374e-01
  -1.09715271e+00 -8.88255715e-01  6.76214874e-01  1.26331055e+00
   9.71547604e-01  1.26690733e+00  2.13557646e-01  7.49732792e-01
   9.17353332e-01 -1.35699183e-01  1.02075779e+00  1.39578366e+00
  -3.95577759e-01  1.82343590e+00  1.22511053e+00  1.37567306e+00
   6.19295895e-01 -4.07072157e-01 -3.15917075e-01  4.38034356e-01
   2.15873435e-01 -1.61093962e+00 -2.36747146e+00  1.51434287e-01
  -2.26472473e+00 -1.59191358e+00  1.72776997e-01 -1.72481596e+00
  -3.97043675e-01 -1.37065017e+00  1.18780577e+00 -6.37652636e-01
   3.37333083e-01 -8.84323299e-01  7.01664329e-01 -1.20757654e-01
   6.49434865e-01  6.22194529e-01  6.06806695e-01 -1.30204308e+00
  -7.03575671e-01 -1.22511554e+00  1.10196507e+00 -1.30670393e+00
   4.24225092e-01 -2.37956554......

but when I got it for pretrained

model=models.resnet50(pretrained=True)
........
context.execute_v2(bindings=[int(devide_in), int(devide_out)])

it return false

Start by reproducing the two model-loading paths, comparing models.resnet50(ResNet50_Weights.IMAGENET1K_V1) with models.resnet50(pretrained=True), and record the complete execute_v2 error output. Inspect the bindings passed to context.execute_v2 and the model conversion or engine-building steps mentioned in the report. Done means identifying the cause of the failure or documenting a minimal reproducible case with the required environment details.