RuntimeError：pytorch中的cuda运行时错误（8）

Question

when I try to run a pytorch faster rcnn code (from https://github.com/rowanz/neural-motifs ), I get the issuse as follow

THCudaCheck FAIL file=/opt/conda/conda-bld/pytorch_1518238409320/work/torch/lib/THC/generic/THCTensorMathPairwise.cu 
line=21 error=8 : invalid device function

It arises when operate keep.append(keep_im + s) where keep is a list, s is a int number and keep_im is a torch.cuda.Longtensor

It's strange that, when I modify the code as

try:
    keep.append(keep_im + s)
except BaseException:
    keep.append(keep_im + s)

It return error in the try and then operates again in the except, and successes....

Anyone know what happen here? I use python 2.7 + pytorch 0.3 + Cuda 8 + cudnn7.1, Titan XP in Ubuntu 16. Thanks

Answer 1

You have to upgrade torch version. **

For me upgrade to torch 1.11.0 version worked.

**