predict fault and server die

when i run code that here :
self.predictor = DefaultPredictor(cfg)
predictions = self.predictor(image)

Environment:
detectron2 0.5+cu110 pypi_0 pypi
pytorch 1.7.1 py3.8_cuda11.0.221_cudnn8.0.5_0 pytorch
pytorch-lightning 1.2.1 pypi_0 pypi
cudatoolkit 11.0.221 h6bb024c_0 defaults
cudnn 8.0.5.39 ha5ca753_1 conda-forge

MY GPU :
Tesla P100-PCIE 16G

Logs:

--------------------------------------
C++ Traceback (most recent call last):
--------------------------------------
0 at::Tensor::addmm(at::Tensor const&, at::Tensor const&, c10::Scalar, c10::Scalar) const
1 at::addmm(at::Tensor const&, at::Tensor const&, at::Tensor const&, c10::Scalar, c10::Scalar)
2 at::native::addmm_cuda(at::Tensor const&, at::Tensor const&, at::Tensor const&, c10::Scalar, c10::Scalar)
3 at::native::addmm_out_cuda(at::Tensor&, at::Tensor const&, at::Tensor const&, at::Tensor const&, c10::Scalar, c10::Scalar)
4 at::native::(anonymous namespace)::addmm_out_cuda_impl(at::Tensor&, at::Tensor const&, at::Tensor const&, at::Tensor const&, c10::Scalar, c10::Scalar)

----------------------
Error Message Summary:
----------------------
FatalError: `Segmentation fault` is detected by the operating system.
[TimeInfo: *** Aborted at 1658132929 (unix time) try "date -d @1658132929" if you are using GNU date ***]
[SignalInfo: *** SIGSEGV (@0x0) received by PID 8476 (TID 0x7f0e017fa700) from PID 0 ***]