Double Speed for forcing fp32 with Zluda on flux

[DeeH0pe]

Anyone else noticed a speed boost while consuming more vram with this commandline args? Model: GGUF Q4 --use-zluda --attention-quad --all-in-fp32

Im about around 3s/it faster than with bfloat16

[mirh]

Yes, you aren't the first one https://github.com/lllyasviel/stable-diffusion-webui-forge/issues/1684

[VeteranXT]

You can use Q4? How?

[DeeH0pe]

You can use Q4? How?

Just put the Q4 in Stable-diffusion folder. Open launch.py and add these 2 lines and adjust your Path to the packages_3rdparty folder:

import sys sys.path.append(r"Your_Path\packages_3rdparty")

[bigwinboy]

You can use Q4? How?

Just put the Q4 in Stable-diffusion folder. Open launch.py and add these 2 lines and adjust your Path to the packages_3rdparty folder:

import sys sys.path.append(r"Your_Path\packages_3rdparty")

Hello, How is the sampler etc set up please, I am able to run it but all that comes out is a black picture.

[mirh]

Black picture may be https://github.com/lllyasviel/stable-diffusion-webui-forge/issues/1278

[bigwinboy]

Black picture may be lllyasviel#1278

Thank you, successfully generated images after adding the command --all-in-fp32 to COMMANDLINE_ARGS= in webui-user.bat.