getting "For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions." about magic-animate-for-windows

[sdbds]

your torch with cuda 118,but your install cuda 12.1 in locally...
plz change requiments-windows about --extra-index-url https://download.pytorch.org/whl/cu118
to --extra-index-url https://download.pytorch.org/whl/cu121

[swpatrick]

C:\Users\PatZ>nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Wed_Sep_21_10:41:10_Pacific_Daylight_Time_2022
Cuda compilation tools, release 11.8, V11.8.89
Build cuda_11.8.r11.8/compiler.31833905_0

C:\Users\PatZ>time
The current time is: 16:53:20.28
Enter the new time:
C:\Users\PatZ>date
The current date is: 12/12/2023 Tue
Enter the new date: (mm-dd-yy)

should be this, I didnt reopen the cmd window after i install the 11.8 cuda, sorry.
I had 12.1 at first

[sdbds]

C:\Users\PatZ>nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Wed_Sep_21_10:41:10_Pacific_Daylight_Time_2022
Cuda compilation tools, release 11.8, V11.8.89
Build cuda_11.8.r11.8/compiler.31833905_0

C:\Users\PatZ>time
The current time is: 16:53:20.28
Enter the new time:
C:\Users\PatZ>date
The current date is: 12/12/2023 Tue
Enter the new date: (mm-dd-yy)

should be this, I didnt reopen the cmd window after i install the 11.8 cuda, sorry.

we recommend python 3.10 because some deps happend errors in 3.11

[swpatrick]

C:\Users\PatZ>python --version
Python 3.10.11

C:\Users\PatZ>time
The current time is: 17:23:53.55
Enter the new time:

C:\Users\PatZ>date
The current date is: 12/12/2023 Tue
Enter the new date: (mm-dd-yy)

C:\Users\PatZ>

--extra-index-url https://download.pytorch.org/whl/cu118/torch-2.0.1%2Bcu118-cp310-cp310-win_amd64.whl

I installed python 3.10 but got same error

[sdbds]

if your local cuda is 12.1，plz install cu121.
it is cuda environment problems...

[swpatrick]

what do you mean local cuda? , isnt cuda version is depend on what CUDA Toolkit you have installed on you machine? I had cuda 11.1 at the begining(from nvidia control pannel - system info), then i saw the project need 11.3 or above, so i install the 12.3 version, but torch does not support the 12.3 yet so I installed 12.1, sooner I figured out the program require v0.15 torchvision which only gonna work with v2.0 torch, and v2.0 torch does not support cuda 12.1 so finally I installed 11.8.

[sdbds]

check your environment path in windows, if you have multi cuda, it will use lastest version.

[sdbds]

what do you mean local cuda? , isnt cuda version is depend on what CUDA Toolkit you have installed on you machine? I had cuda 11.1 at the begining(from nvidia control pannel - system info), then i saw the project need 11.3 or above, so i install the 12.3 version, but torch does not support the 12.3 yet so I installed 12.1, sooner I figured out the program require v0.15 torchvision which only gonna work with v2.0 torch, and v2.0 torch does not support cuda 12.1 so finally I installed 11.8.

we use venv torch with cuda version like cu118 or cu121
but your system should have a cuda installed.

[swpatrick]

this is my enviorment path and use pip list after venv/script/active followed with nvcc --version

[sdbds]

this is my enviorment path and use pip list after venv/script/active followed with nvcc --version

as i known, GTX980 use CC with SM5.2
I guess cuda 11.8 or xformers ==0.0.22 deprecated SM5.2

(xformers 0.0.22 maybe)

You can down level these version for your GPU gtx980