Need help RuntimeError: CUDA error: device-side assert triggered

1

Hii, I’ have made a Graph convolution network which takes 2 graphs as inputs and gives output between 0 to 1 , After running on one sample , it shows this error . Please help

image
image1089×518 61.6 KB

2

The error points to an invalid index.
You could rerun the code either on the CPU to get the line of code causing the issue or on the GPU via CUDA_LAUNCH_BLOCKING=1 as suggested in the error message.

3

Thanks @ptrblck . I was working with graphs and after trying to run on cpu ,i found that there was a mismatch between number of nodes (N) and shape of adjacency matrix which in ideal case should be N x N .