如何解决Tile中Tensorflow的分段错误？

Question

我正在运行这段代码（https://github.com/wanderine/ProgressiveGAN3D），并得到半随机分段错误。我用“gdb”来定位分段错误来自Tensorflow中的“tile”函数，但我很不知道从这里开始该怎么做，因为从Tensorflow图内部打印不是直接的。我怀疑TF在执行平铺操作时会耗尽GPU内存，但即使在大量减少批处理大小后，我也会遇到相同的错误

#5  0x00007ffe1ef89a03 in ?? () from /usr/lib/x86_64-linux-gnu/libcuda.so.1
#6  0x00007ffe1f10778b in cuMemcpyHtoDAsync_v2 () from /usr/lib/x86_64-linux-gnu/libcuda.so.1
#7  0x00007fffaee10dcc in cudart::driverHelper::memcpyAsyncDispatch(void*, void const*, unsigned long, cudaMemcpyKind, CUstream_st*, bool) ()
   from /home/andek67/.conda/envs/mypgan3d/lib/python3.7/site-packages/tensorflow/python/_pywrap_tensorflow_internal.so
#8  0x00007fffaede9f13 in cudart::cudaApiMemcpyAsync(void*, void const*, unsigned long, cudaMemcpyKind, CUstream_st*) ()
   from /home/andek67/.conda/envs/mypgan3d/lib/python3.7/site-packages/tensorflow/python/_pywrap_tensorflow_internal.so
#9  0x00007fffaee2c5a8 in ?? ()
   from /home/andek67/.conda/envs/mypgan3d/lib/python3.7/site-packages/tensorflow/python/_pywrap_tensorflow_internal.so
#10 0x00007fffa966a8df in void tensorflow::internal::TileSimple<float>(Eigen::GpuDevice const&, tensorflow::Tensor*, tensorflow::Tensor const&) ()
from /home/andek67/.conda/envs/mypgan3d/lib/python3.7/site-packages/tensorflow/python/_pywrap_tensorflow_internal.so
#11 0x00007fffa966b08d in tensorflow::functor::Tile<Eigen::GpuDevice, float, int>::operator()(Eigen::GpuDevice const&, tensorflow::Tensor*, tensorflow::Tensor const&, absl::Span<int const>) const ()
``