Billy Cao
Wait you cannot even use it with different cuda versions: ``` Traceback (most recent call last): File "E:\PyCharmProjects\stable-diffusion-webui\launch.py", line 298, in start() File "E:\PyCharmProjects\stable-diffusion-webui\launch.py", line 289, in start import webui...
My draft PR includes a workaround for the codeformer warning. Its no big deal
As the error message suggests, the training script expects torch v1 but you are running torch v2. This cannot be fixed until the upstream repo (SD repo by stability-ai) migrates...
@Thomas-MMJ I have not encountered this error myself so sorry about the misread. Let's wait and see if anyone else report the same and maybe we can find out the...
@Panchovix could you try to remove the `torch.compile` line and see if the error is still there?
So this is not a torch.compile issue but maybe a torch 2.0 issue.
Interesting. Lets see if anyone else with 4090 have the same issue.
> I've got bad news for you 2.0 testers: > > https://pytorch.org/blog/compromised-nightly-dependency/ Only for linux users
You do not have to use WSL. In fact, using it can bring additional overhead. It is already surprising that you did not get a performance loss. As mentioned in...
@Zuxier @Panchvzluck Try if https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/6465 fixes your training issue