Doesn't generate anything on 5090 #7
Comments
jim60105
added a commit
that referenced
this issue
Feb 15, 2025
Update Dockerfile and entrypoint.sh to support new graphic cards with updated CUDA version and xformers environment variable. #7 * **Dockerfile** - Update the CUDA version to 128 in the `pip install` command. - Update the `torch`, `torchvision` versions to the latest nightly builds and `xformers` to `0.0.29.post2`. * **entrypoint.sh** - Update the `pip install` command to use the latest nightly builds of `torch`, `torchvision`, and `xformers==0.0.29.post2`.
jim60105
added a commit
that referenced
this issue
Feb 15, 2025
Update Dockerfile and entrypoint.sh to support new graphic cards with updated CUDA version and xformers environment variable. #7 * **Dockerfile** - Update the CUDA version to 128 in the `pip install` command. - Update the `torch`, `torchvision` versions to the latest nightly builds and `xformers` to `0.0.29.post2`. * **entrypoint.sh** - Update the `pip install` command to use the latest nightly builds of `torch`, `torchvision`, and `xformers==0.0.29.post2`. Signed-off-by: CHEN, CHUN <[email protected]>
|
Please check if these two images work on your 5090
(Note that they are temporary tags for testing and will be removed after #8 merge into master.) |
|
So I updated the forge docker compose with the new tag, enabled --reninstall-torch in the commands section the compose file, but now the container fails to start. Did I mess a step or something? |
|
Do not reinstall torch and xformers, it is the purpose of this issue. And if you are using the |
|
I'm pretty much using your default yaml files I usually use forge, I tried again with the base a1111 file and got the same result. |
|
Please try and then to ensure you use the correct container image 🤔 |
I got the same error sadly. I also tried going inside the container and running since the xformers are updated. But I got dependecies issue EDIT: I used protobuf 4.25.6 "didn't try latest" and mediapipe 0.10.21 and open-clip-torch 2.30.0 |
jim60105
added a commit
that referenced
this issue
Feb 20, 2025
Update Dockerfile and entrypoint.sh to support new graphic cards with updated CUDA version and xformers environment variable. #7 * **Dockerfile** - Update the CUDA version to 128 in the `pip install` command. - Update the `torch`, `torchvision` versions to the latest nightly builds and `xformers` to `0.0.29.post2`. * **entrypoint.sh** - Update the `pip install` command to use the latest nightly builds of `torch`, `torchvision`, and `xformers==0.0.29.post2`. Signed-off-by: CHEN, CHUN <[email protected]>
|
You are correct. I just noticed in my docker build log that it initially collected torch 2.7.0, but actually used torch 2.6.0 from the default source after collected xformers. I found this issue pointing out that we need to build it from source at the moment, and this is already a bit too far from this project. However considering other users outside the Blackwell series, I believe xformers should not be removed from my default images. |
I have fixed these two images to use torch 2.7.0 If you still get the same pip dependency error, it is very likely a configuration issue with your container rather than a problem with the docker image itself.
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
It seems the new graphic cards don't work on the current installed cuda version
Take a look here
lllyasviel/stable-diffusion-webui-forge#2601
I tried running the update command directly inside the container
Now I get another error about xformers need to be updated or something
I think we need to set env variable for the xformers here
like mentioned here AUTOMATIC1111/stable-diffusion-webui#16818
but now sure how for your image.
The text was updated successfully, but these errors were encountered: