Get green screen With my 1660 SUPER. (ft16 issue)

[UrielCh]

Hi,

It looks like my 1660 SUPER does not support ft16 operation, is it possible to add a parameter to switch to full precision?

thx.

[AUTOMATIC1111]

--precision full is there

[UrielCh]

I tryed to change start line in relauncher.py to

os.system("python scripts/webui.py --precision full")

after ensuring the presence of:

• model.ckpt
• GFPGANv1.3.pth
• RealESRGAN_x4plus.pth
• RealESRGAN_x4plus_anime_6B.pth

and a new Test for a 128x128 Img, I foubnd in my browser console:

CRASHED:<br><textarea rows=\"5\" style=\"color:white;background: black;width: -webkit-fill-available;font-family: monospace;font-size: small;font-weight: bold;\">expected scalar type Half but found Float</textarea><br><br>Please wait while the program restarts."

type Half but found Float...

yep... expected fp16 but get fp32? that is the purpose of this test ...

[orionaskatu]

You seem to be using the hlky fork from here : https://github.com/hlky/stable-diffusion-webui

Found on the installation guide (https://rentry.org/GUItard) :

If your output is a solid green square (known problem on GTX 16xx):
Add --precision full --no-half to the launch parameters above, it should look like this:
"python scripts/webui.py --precision full --no-half --optimized"
Unfortunately, the full precision fix raises ram use drastically so you may have to moderately reduce your output to 448x448 if on 4gb

[UrielCh]

Thx that better with --precision full --no-half.
Now I need a new GFX, 6GB in f32 is... limiter

I did not see this line under RUNNING ON 4GB section My GFX is 6GB

[ZeroCool22]

Where is realuncher.py?

i can't find it...

[orionaskatu]

Where is realuncher.py?

i can't find it...

There is no relauncher.py in this repo, parameters have to be passed directly to webui.py

[ZeroCool22]

Where is realuncher.py?
i can't find it...

There is no relauncher.py in this repo, parameters have to be passed directly to webui.py

Can you teach me the correct way to do it please?

I just copy the one i want from here: https://github.com/hlky/stable-diffusion/blob/d667ff52a36b4e79526f01555bfbf85428f334ce/scripts/webui.py

And add it like this?:

Or i need to add some more lines?

[orionaskatu]

It won't work that way, you don't have to modify the content of .py files.
You just have to launch webui.py from command line with the wanted parameters like:
python webui.py --precision full --no-half

Also don't mix files from the hlky repo and this one, it won't work as they are very different.

[ZeroCool22]

It won't work that way, you don't have to modify the content of .py files. You just have to launch webui.py from command line with the wanted parameters like: python webui.py --precision full --no-half

Also don't mix files from the hlky repo and this one, it won't work as they are very different.

Ok, very thx for the info, i always used the GUI'S ones, so i don't know much about the correct way use of command lines.

Will try it.

[ZeroCool22]

It won't work that way, you don't have to modify the content of .py files. You just have to launch webui.py from command line with the wanted parameters like: python webui.py --precision full --no-half

Also don't mix files from the hlky repo and this one, it won't work as they are very different.

Trough Anaconda:

And DIRECT from cmd:

When i execute the webui.bat everything runs fine, but fro mthe console get that error.

[orionaskatu]

Seems you're missing dependencies in conda and your default python installation.
They are automatically installed in a virtual env by the webui.bat

The easiest way for you is then to change line 5 of webui.bat to:
set COMMANDLINE_ARGS=--precision full --no-half

And then it should work by launching webui.bat

[ZeroCool22]

It was telling me always something was missing, so i run pip install -r requirements.txt, and now it start with no problems.

[AmmBe]

I installed the PyTorch 1.12.1 cuda 11.6 and extracted the cudnn 8.5 archive to the \torch\lib folder. Now I can generate without using --precision full, but only at 320 and 640 resolutions and the speed is twice as slow.