Fresh install, issue found and resolved, issue yet to resolve

[eroyee]

@simonmcnair good on you for carrying on development of this 👍

FYI I've had Stablediff/Auto1111 running on Ubuntu 24.04 in the past but it's been a mare trying to deal with the Python version required for that alongside other things running on the machine so I thought I'd see if there was a Docker option, and ended up here today.

I went through the setup detail here, using your repo, but had a few hiccups which I thought might be worth documenting in case anyone else has a similar problem. First issue was:

Error response from daemon: could not select device driver "nvidia" with capabilities: [[compute utility]]

My bad, I have an elderly 1070 in the machine so needed to follow this, then systemctl restart docker...

Which led to the next issue:

........ auto-1 | Downloading: "https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned-emaonly.safetensors" to /stable-diffusion-webui/models/Stable-diffusion/v1-5-pruned-emaonly.safetensors
auto-1 |
auto-1 | loading stable diffusion model: FileNotFoundError........

I fixed this by installing v1-5-pruned-emaonly.safetensors from the existing baremetal install into data/models/Stable-diffusion/ obviously downloading a checkpoint from elsewhere would have achieved the same result. Useful info here.

However I'm now at a point where it gives:

....OSError: Can't load tokenizer for 'openai/clip-vit-large-patch14'....

There seem to be a few responses to this, for example here and here, but I'm not sure about the file hierarchy of the docker image and exactly what's exposed in the volumes so wonder if you have any advice on this?

Thanks.

[eroyee]

Just had a moment to get back to this. Looking at the running docker instance I see it appears to have models/Stable-diffusion/v1-5-pruned-emaonly.safetensors in place so perhaps there's something awry with the declared paths?

Also, making the host data directory that of the existing baremetal install did not fix the tokeniser issue.

Out of interest I'm presently d/loading the forge install but am on a ridiculously slow data connection so it'll be a while before that completes...

UPDATE: Well it works successfully using forge HOWEVER when I tried installing a couple of extensions (deoldify and old-photo-restoration) it broke, quite badly and is not presently recoverable. I've had a basic go at trying to fix the issues but haven't succeeded, happy to discuss further and post logs etc if anyone is interested...

[eroyee]

Further update: It appears the issues I was having with the Forge UI are due to the extensions themselves. A baremetal install revealed exactly the same problems with those extensions, so as far as I'm aware the docker instance [of Forge] is operating fine.

[simonmcnair]

Sorry for the delay. I have been thinking about the download issue and think I might have a fix.

[eroyee]

No hurry, was just providing some feedback.

I haven't had it running since I posted last but as posted I seem to recall seeing the checkpoint file present at one stage when I logged into the running docker instance. Not sure now if that was actually the external volume file, or whether it was already present in the image and just not seeing it for some reason? Could well be just me getting something wrong.

While that was fixable by manually installing a checkpoint file the tokeniser thing seemed more problematic at the time. After the Forge instance worked I didn't go back to Auto1111, and now as a result of trying to get to the bottom of the extension issues I'm running a baremetal Forge install. Quite happy to trial your docker work again if it helps, I think it's a worthwhile project, it just takes me a while to do 'cos of the poor bandwidth here.