CUDA-BEVFusion: difference in the result of onnx and pytorch model

[Jayden9912]

Hi.

I have converted BEVFusion model from torch to onnx but it fails np.testing.allclose

Mismatched elements: 17068436 / 100663296 (17%)
Max absolute difference: 0.00343999
Max relative difference: 17035.555
 x: array([[[[0.401343, 0.140742, 0.282421, ..., 1.049651, 1.047727,
          0.911273],
         [0.486538, 0.307632, 0.1848  , ..., 0.661901, 0.757838,...
 y: array([[[[0.401524, 0.140862, 0.282235, ..., 1.049505, 1.047449,
          0.910834],
         [0.486407, 0.307709, 0.18471 , ..., 0.661574, 0.757754,..

Is it normal?