Expected all tensors to be on the same device
Hey there,
I'm creating pods to use Stable Diffusion WebUI, using the tempates offered by RunPod.
In both templates, if I use any model that's not default, I get the following error:
RuntimeError: Expected all tensors to be on the same device, but found at least two
devices, cpu and cuda:0! (when checking argument for argument index in method
wrapper_CUDA index select)
Any ideas on this? There's not much helpful source about this issue.
Thanks
34 Replies
Which templates? There are multiple WebUI templates. Also, which model and how did you download it? Sounds like your model may be corrupt or something.
@ashleyk thanks for the answer. I've tried
runpod/stable-diffusion:web-ui-10.2.1
runpod/stable-diffusion:fast-stable-diffusion-2.4.0
Both had same exact issues.
Which model?
In a fresh installation I can reproduce this error with sd_xl_base_1.0.safetensors [31e35c80fc]
Did you set your resolution to 1024x1024?
I also installed some extra models, if I select them, the webui switches back to sd_xl_base automatically.
Just tried now. same issue
What kind of pod is this?
RTX A5000. I can run these models locally with no problems.
I don't know why this is happening, everything looks okay to me. @Madiator2011 [EU] any idea?
Isn't it reproducable for others?
I created so many pods in past a few days, every one of them had this issue
corupted model
I also suspected that but the SDXL base model is included with the template so it shouldn't be corrupt, maybe some issue with the storage in SE or something thats causing the models to become corrupt.
Would need to see full error messages
If there's a corrupt model in the same folder, would we get the same error, for an uncorrupt model?
No
I see that my uploads have been timing out.
Which is another problem. I'm not sure how to download large models from civitai, which requires login for download.
Maybe the corrupt model messed up the tensors and you have to restart your pod to clear them
Try restarting your pod and then using the SDXL base model again
I created another pod already
Use the CivitAI extension, you have to pay for it though
any reliable way to upload? why does scp time out so often?
You can try runpodctl https://github.com/runpod/runpodctl
GitHub
GitHub - runpod/runpodctl: 🧰 | RunPod CLI for pod management
🧰 | RunPod CLI for pod management. Contribute to runpod/runpodctl development by creating an account on GitHub.
Installation fails
Don't use homebrew
Oh I see they updated the instructions
Do you have Intel or Apple Silicon?
M1
the wget way also does not work
this worked, thanks!
I'll try uploading now
@Justin Merrell homebrew instructions for runpodctl for Mac are not working
I also don't get how file transfer works.
send command doesn't specify a pod name. It's sending a file. but to where?
there seems to be a lot of issues with the runpodctl
I have no clue what's going on here... it zips, then gives a code, files are getting uploaded somewhere and it also says "On the other computer run"
It's very poetic output, but a simple file upload leaves the user cluelessYou don't need the pod name, you type the recieve command that it gives you into your pod to receive the data
You can also download from api directly
Note you need generate api key on civatic
example for Ultraspice would be
@snowmonkey the api key you can get from https://civitai.com/user/account
Civitai: The Home of Open-Source Generative AI
Explore thousands of high-quality Stable Diffusion models, share your AI-generated art, and engage with a vibrant community of creators