Hi , Would you release your experiments environment?

Question

Hi , Would you release your experiments environment?

Johnson-yue opened this issue 4 years ago · comments

I am using, python3.7.3, cuda 10.0, cudnn 7.6.1 and then I run finetune.py as command:
python finetune.py --name AnimalFace_finetune --mixing --loss r1 --sched --dataset AnimalFace

and the Bug is :

cuDNN error: CUDNN_STATUS_NOT_SUPPORTED. This error may appear if you passed in a non-contiguous input

bug occur in line 312 in finetune.py

Do you understand this bug meaning???

Johnson yue · Answer 1 · Thu Mar 05 2020 10:07:23 GMT+0800 (China Standard Time)

Sorry this is environment bug, and I fixed with:

torch : 1.2.0
torchvision : 0.3.0
cuda 10

Sangwoo Mo · Answer 2 · Thu Mar 05 2020 14:24:28 GMT+0800 (China Standard Time)

Hi, sorry for late response! Glad to hear that you fixed it! :)

LendelTheGreat · Answer 3 · Sat Mar 21 2020 01:10:15 GMT+0800 (China Standard Time)

Ran into the same issue. Managed to get it working with a nightly pytorch build:

torch: 1.5.0
torchvision: 0.6.0
cuda 9.1

Johnson yue · Answer 4 · Mon Mar 23 2020 10:16:58 GMT+0800 (China Standard Time)

@LendelTheGreat Thank you

Sangwoo Mo · Answer 5 · Mon Mar 23 2020 11:29:32 GMT+0800 (China Standard Time)

Hi, sorry for the late response!
I remember that I had the same issue, but I forgot it since I fixed it a while ago.

It is a bug in PyTorch 1.4.0 (when one using .transpose() or .permute()), and one can fix it by upgrading to nightly version or downgrading. Thanks for the reminder @LendelTheGreat !