part_seg problem

[zhangdongwei1998]

RuntimeError: CUDA error: CUBLAS_STATUS_NOT_INITIALIZED when calling cublasCreate(handle)

[qq456cvb]

Could you tell me about your Pytorch and CUDA/cudnn version, this error seem to be related to version mismatch or incorrect pytorch gpu installation. You may try to install Pytorch in a clean Anaconda environment.

[zhangdongwei1998]

Thank you for your reply. I'll check it later.

[Confusedisprecious]

Thank you for your reply. I'll check it later.
I had the same issue, how did you end up fixing this issue?

[qq456cvb]

This is almost always either a PyTorch/CUDA driver mismatch or an out-of-memory that shows up as a cuBLAS error. Things to try: install a torch build that matches your CUDA, lower batch_size, make sure gpu points at a valid device (it now defaults to 0), and rerun with CUDA_LAUNCH_BLOCKING=1 to get the real error. If it's OOM, a smaller batch usually sorts it out.