Runtime Error while training

[hashirupramuditha]

When I ran "bash script/train.sh mipnerf360/counter 2" to train the counter dataset, I faced this issue.

"RuntimeError: CUDA error: CUBLAS_STATUS_INVALID_VALUE when calling cublasSgemmStridedBatched( handle, opa, opb, m, n, k, &alpha, a, lda, stridea, b, ldb, strideb, &beta, c, ldc, stridec, num_batches)"

Can you tell me how to resolve this issue?

[Nieschlafen]

you can try

unset LD_LIBRARY_PATH
The reason for the environment variable, this is very likely, because of the different virtual environment settings and the operation of setting the environment variable when installing CUDA and CUDNN at the beginning, so it is very likely for a server used by multiple users