CODEWITHSUNDEEP
stable-diffusion
吾发克
吾发克

Reputation: 261

I Run stable diffusion,It`s wrong RuntimeError: "LayerNormKernelImpl" not implemented for 'Half',

I Run stable diffusion,It`s wrong RuntimeError: "LayerNormKernelImpl" not implemented for 'Half',help me,how i do?

I have no idea for it

Upvotes: 25

Views: 70571

Answers (8)

saqib diar
saqib diar

Reputation: 87

I have faced the same problem, and here is the solution, first do this inside your launch.py:

commandline_args = os.environ.get('COMMANDLINE_ARGS', "")
commandline_args = os.environ.get('COMMANDLINE_ARGS', "--skip-torch-cuda-test --no-half")

If you have done it already and still not working then you need to run the below command

./webui.sh --precision full --no-half

Upvotes: 0

Nestor Ledon
Nestor Ledon

Reputation: 1992

In newer versions of the A1111 WebUI you can add the needed arguments to webui-user.sh

# Commandline arguments for webui.py, for example: export COMMANDLINE_ARGS="--medvram --opt-split-attention"
export COMMANDLINE_ARGS="--skip-torch-cuda-test --no-half"

Upvotes: 1

Guide Studio
Guide Studio

Reputation: 41

RuntimeError: “LayerNormKernelImpl” not implemented for ‘Half’

I have successfully solved this problem by way of

Start the webUI with the following Mac OS Terminal command.

cd stable-diffusion-webui ./webui.sh --precision full --no-half

Special Thank >> https://stable-diffusion-art.com/install-mac/comment-page-1/

Upvotes: 4

Simon TAMAYO
Simon TAMAYO

Reputation: 91

Instantiate the pipeline with float32, it will run slower but it will run ;)

pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float32)

Upvotes: 9

Jason Forrest
Jason Forrest

Reputation: 59

In Windows System, Edit webui-user.bat file

set COMMANDLINE_ARGS= --skip-torch-cuda-test --no-half

Upvotes: 5

Manoj Mohan
Manoj Mohan

Reputation: 6044

'LayerNorm' is one of the layers in the Model. Looks like you're trying to load the diffusion model in float16(Half) format on CPU which is not supported. For float16 format, GPU needs to be used. For CPU run the model in float32 format.

Reference: https://github.com/pytorch/pytorch/issues/52291

Upvotes: 30

salvador rios
salvador rios

Reputation: 191

I solved it in the following way, modify the launch.py file in the following way, add the following line of code:

commandline_args = os.environ.get('COMMANDLINE_ARGS', "--skip-torch-cuda-test --no-half")

launch.py file and the modified line of code just as stated above

Upvotes: 19

Rains
Rains

Reputation: 111

I am not sure what your environment is. I can show you what I have encountered. I also encountered this error during my Colab testing, which appeared after I changed my model. Then I realized that I had forgotten to assign the pipe to CUDA. After running the code below, everything was fine.

pipe = pipe.to("cuda")

Upvotes: 11

Related Questions