search

AUTOMATIC1111 / stable-diffusion-webui

Stable Diffusion web UI
GNU Affero General Public License v3.0
139.22k stars 26.42k forks source link

[Bug]: Inactive maintainer #9888

Closed jlsnow301 closed 1 year ago

jlsnow301 commented 1 year ago

Is there an existing issue for this?

What happened?

The maintainer of stable-diffusion-webui has been inactive for a month.

Screenshot 2023-04-26 101318

Steps to reproduce the problem

  1. Write a PR.
  2. Wait.
  3. ...?

What should have happened?

Contributions should have been reviewed and merged on a semi-regular basis

Even GitHub is starting to worry: Screenshot 2023-04-26 101215

Commit where the problem happens

22bcc7b

What platforms do you use to access the UI ?

Windows

What browsers do you use to access the UI ?

Brave

Command Line Arguments

No

List of extensions

Default

Console logs

None

Additional information

I really hope this project can pick up steam again. No ill will towards automatic1111. The lack of a maintainer presents an existential threat to the codebase.

DutchMike64 commented 1 year ago

Sadly AMD support broke some time ago and is very difficult to get working at the moment. I hope the creator or someone with a love for AMD GPU's makes it easy again.

midcoastal commented 1 year ago

What does AMD have to to with anything? Certainly, great AMD support would be awesome, especially considering the vRAM to $ ratio, but this isn't about any GPU support, it is about the developer literally ghosting (from Git and Redit as a whole).

Just straight vanished. Does anyone know of any other way of communicating with them? Like: are they alive, at least?

In case anyone is wondering, there are some active forks:

https://github.com/anapnoe/stable-diffusion-webui-ux https://github.com/vladmandic/automatic

But, there is no longer a central repo, which is unfortunate...

Someone REALLY needs to consolidate this project again. Not even a sole developer, just someone to gather the troops/effort, as opposed to a slew of individuals just forking and going it on their own... :-/

jlsnow301 commented 1 year ago

Like: are they alive, at least?

I'm hoping they're okay. This is a pretty big project for someone to just disappear on. Really makes you wonder.

https://github.com/vladmandic/automatic looks promising.

DutchMike64 commented 1 year ago

What does AMD have to to with anything? <

You misunderstood my remark : Automatic1111 used to work with AMD Gpu's without a problem, now it doesn't.

carco5a commented 1 year ago

https://github.com/vladmandic/automatic looks promising.

A second for this repo. Dev is engaged and active for now at least. Pretty easy to symlink/junction folders from A1111 into this tree to avoid duplication if needed.

midcoastal commented 1 year ago

What does AMD have to to with anything? <

You misunderstood my remark : Automatic1111 used to work with AMD Gpu's without a problem, now it doesn't.

I think you misunderstood my confusion: There are ~2k outstanding issues on this repo right now, AMD support is so wildly left-field in the scope of this conversation. I get it, AMD support is high on your priority list, and it would be nice if the dev came back to fix it, and I don't fault you for that.

The confusion came about because the more pertinent topic of this particular conversation is less about "oh, I hope they come back to fix this particular thing," and more "well shoot, this project may be totally dead in the water, along with all of the, literally thousands of, bugs... darn"

Again, not giving you a hard time, I just can't get bent out of shape about one or two things that are broken when it looks like it's straight up abandoned.

A month with no updates? Whelp... moving on, I suppose.

If you read this, @AUTOMATIC1111: Thank you for everything you have done. Tremendous respect. I hope you are well, I really do. That said, the industry is moving incredibly fast, and we've got to move on. When you come back, maybe pick the best few forks/authors, and get them all to work together.

Until then, cheers. Thanks again!

DutchMike64 commented 1 year ago

A month with no updates? Whelp... moving on, I suppose. Until then, cheers. Thanks again!

I see now :-) and agree with your final statement, thanks to automatic1111 for bringing AI image generation to many. Hope someone picks it up and meanwhile.... Maybe for this is should trade to a nvidia....

DGdev91 commented 1 year ago

Sadly AMD support broke some time ago and is very difficult to get working at the moment. I hope the creator or someone with a love for AMD GPU's makes it easy again.

i wrote a PR for that, wich of course has never been merged. #9404

YichenDongBrex commented 1 year ago

Wait what's wrong with AMD? I've been using it just fine, but I'm not sure when the last time I pulled from this repo. I did make this as the torch command in the webui.sh file though: pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm5.4.2. Also I have a 6900xt in case it's a GPU difference

DutchMike64 commented 1 year ago

Wait what's wrong with AMD? I've been using it just fine, but I'm not sure when the last time I pulled from this repo.

Exactly, it used to work without a hiccup and then suddenly it didn't. I also replaced the pip3 command but at the moment nothing works anymore with a new install. I did get comfyui to work though. My GPU is a 6700XT but like i said, it worked before.

DGdev91 commented 1 year ago

Wait what's wrong with AMD? I've been using it just fine, but I'm not sure when the last time I pulled from this repo. I did make this as the torch command in the webui.sh file though: pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm5.4.2. Also I have a 6900xt in case it's a GPU difference

The problem is that the old command used --extra-index-url instead of --index-url. The problem went out the day when pytorch2 was released. It finds that the main repository has a newer version than the one in the old AMD repo, do it tries to install that. ....Wich is made for nvidia

The PR #9404 fixes that.

...Or you can just install the ROCm version of pythorch2, just as you did. (But seems like it has some issues with older gpus like the 5700XT)

midcoastal commented 1 year ago

...Or you can just install the ROCm version of pythorch2, just as you did. (But seems like it has some issues with older gpus like the 5700XT)

Or we could just move on and pick a new repo to put our energy into. I have personally tried both of the above mentioned repos, but prefer https://github.com/anapnoe/stable-diffusion-webui-ux

I like the idea that Vladmandic isn't afraid of really mixing things up at the core, and the CLI output is nice, but I find that the end-effect as a USER of the Anapnoe repo is much greater.

Vladmandic has done some UI stuff as well, but the experience is just so much better with Anapnoe. I really enjoy it.

If only they could... combine???

FFS, can we just get a good group together already? We have brilliant minds all over this, why can't someone just go buy a damn domain, contact all these super smart people, and get it all under one roof, and not spread out all willy nilly like this.

It would mitigate this issue in the future... Someone? Anyone? I have some infrastructure, but absolutely zero time to commit...

fbellgr commented 1 year ago

Exactly, it used to work without a hiccup and then suddenly it didn't.

Suddenly? There hasn't been any change for a while.

My RX 6750 XT works fine in Linux (as well as AMD can work with this in terms of performance).

I personally switched to https://download.pytorch.org/whl/nightly/rocm5.4.2/ at some point.

git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui
cd stable-diffusion-webui
python -m venv venv
source venv/bin/activate
pip install --upgrade pip wheel
export HSA_OVERRIDE_GFX_VERSION=10.3.0
export TORCH_COMMAND='pip install torch torchvision --pre --upgrade \ 
        --extra-index-url https://download.pytorch.org/whl/nightly/rocm5.4.2'
python launch.py --skip-torch-cuda-test \
        --opt-sub-quad-attention --upcast-sampling \
        --no-half --disable-nan-check

Remember this is for MY card model.

missionfloyd commented 1 year ago

FFS, can we just get a good group together already? We have brilliant minds all over this, why can't someone just go buy a damn domain, contact all these super smart people, and get it all under one roof, and not spread out all willy nilly like this.

We just need at least one other maintainer.

midcoastal commented 1 year ago

Just going to leave this here: https://github.com/anapnoe/stable-diffusion-webui-ux/issues/75

highnrgappalachian commented 1 year ago

He's back. Chill.

wzgrx commented 1 year ago

My God, the developer is back

jlsnow301 commented 1 year ago

automatic1111 seems to be active again

DutchMike64 commented 1 year ago

Exactly, it used to work without a hiccup and then suddenly it didn't. Suddenly? There hasn't been any change for a while.

Tried your method and for me it uses the CPU instead, i still have an older ubuntu install where everything works, so i'll keep using that in the meantime. But all tries to get anything running on another install are failing. I have an 6700xt which is the same card as you have (6750 only has higher clockspeeds)

fbellgr commented 1 year ago

@DutchMike64

Did you start over from scratch with a new clone, using the venv?

I have Mint 21, which is base on Ubuntu 22.04.

I use the drivers included with the kernel (5.15), NOT those you get through amdgpu-install.

When I started to try to make this work, I had issues with actually getting the GPU involved. I ended up uninstalling pretty much all the packages installed through pip.

When you start messing around without using an environment, sometimes you get stuck. To get it working, it's really best to start fresh and setup the venv properly.

Admittedly, when you have an AMD GPU, you have to learn a few things before it works. But ultimately, there is nothing special to it.

I also installed vladmandic/automatic a few days ago, and right of the bat you have to look at the code to figure out that the TORCH_COMMAND variable is not setup the same as automatic1111.

fbellgr commented 1 year ago

At some point I ran into an error that I fixed by installing libstdc++-12-dev.

Another thing I did (not necessary to make it work, but it's more pleasant) was to build and install CoreCtrl to undervolt the card. Whenever I was stressing it a bit for too long or tried to render a larger image, my computer would hard reset. Using CoreCtrl and the latest ROCm with pythorch 2 seems to have made the resets go away. But again, this is not about making it work, per se.

DutchMike64 commented 1 year ago

@fbellgr

I got stuck with ubuntu 23.04 that didn't work, but got it working today with a fresh install of Arch's EndeavourOS. Today's fresh update and after install i did a "pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm5.4.2" after which everything started to work again :-)

The whole linux thing is highly confusing with amdgpu-install not working on some distro's, in the end i found it it makes things worse instead of better, some distros come with another version of python and installing multiple python versions took me hours, and so on.... Linux is sort of fun, but also very much not... Anyway finally have a running new system now

Thanks to everyone here trying to make stuff work and the hobby would be a lot easier with a Nvidia card, but i don't want to upgrade the 6700xt just for this. (Until nvidia wakes up and starts asking normal prices for a 4070,80,90..)