Closed yishuaidu closed 2 weeks ago
same for me, put the models in the root of /models and to the subdirectories as it would be expected in an automatic1111 installation. Still, I'm getting this error.
Same issue here.
Inside cog.yaml
- python cog-sd-webui/init_env.py --skip-torch-cuda-test
This command is downloading stable-diffusion-stability-ai and other folders inside /stable-diffusion-webui/repositories
but the docker WORKDIR is supposed to be /src
so it cannot find it at /src/repositories/stable-diffusion-stability-ai
try git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui
before you run cog predict ...
Hi @philz1337x, thank you for your work!
I think there are many things missing, we also need to add ControlNET + MultiDiffusion in the extensions/
folder.
The upscaler 4x-UltraSharp.pth in models/ESRGAN
The negative embeddings in /embeddings
We also need to put the correct VAE in models/VAE
(vae-ft-mse-840000-ema-pruned.safetensors)
Basically if we're using CI/CD to push our model on replicate, the current implementation won't work by default, it will require a bit more work.
Would be nice to have it ready to run/deploy by default :)
In the meantime I will try to make it run on my machine and Replicate. Will keep you posted!
Thanks for testing I will extend the readme.
In the meantime try this to run the replicate model on your machine:
docker run -d -p 5000:5000 --gpus=all r8.im/philz1337x/clarity-upscaler@sha256:803dc4af7cdfda701188bd3e009edfd3966ea5c51f9629bc9664befc1829b865
curl -s -X POST \
-H "Content-Type: application/json" \
-d $'{
"input": {
"image": "https://link-to-image",
"prompt": "masterpiece, best quality, highres, <lora:more_details:0.5> <lora:SDXLrender_v2.0:1>",
"dynamic": 6,
"sd_model": "juggernaut_reborn.safetensors [338b85bc4f]",
"creativity": 0.485,
"resemblance": 1.5,
"scale_factor": 2,
"tiling_width": 112,
"tiling_height": 144,
"negative_prompt": "(worst quality, low quality, normal quality:2) JuggernautNegative-neg"
}
}' \
http://localhost:5000/predictions
Thanks for testing I will extend the readme.
In the meantime try this to run the replicate model on your machine:
docker run -d -p 5000:5000 --gpus=all r8.im/philz1337x/clarity-upscaler@sha256:803dc4af7cdfda701188bd3e009edfd3966ea5c51f9629bc9664befc1829b865 curl -s -X POST \ -H "Content-Type: application/json" \ -d $'{ "input": { "image": "https://link-to-image", "prompt": "masterpiece, best quality, highres, <lora:more_details:0.5> <lora:SDXLrender_v2.0:1>", "dynamic": 6, "sd_model": "juggernaut_reborn.safetensors [338b85bc4f]", "creativity": 0.485, "resemblance": 1.5, "scale_factor": 2, "tiling_width": 112, "tiling_height": 144, "negative_prompt": "(worst quality, low quality, normal quality:2) JuggernautNegative-neg" } }' \ http://localhost:5000/predictions
The goal isn't only to run it locally but also to deploy it on replicate as a private/public model with a different config and GPU :)
@philz1337x Do you think you can deploy a copy as a public model with a A100 GPU ? So that people can chose between A40 and A100 :)
@philz1337x Do you think you can deploy a copy as a public model with a A100 GPU ? So that people can chose between A40 and A100 :)
+1, Would love to get an option to run it on A100
Ok I just switched the model to A100 https://twitter.com/philz1337x/status/1772115599726178682
You're the best!
I spent a day deploying it on replicate.com. The day is wasted but still, the same error appears:
AssertionError: Couldn't find Stable Diffusion in any of: ['/src/repositories/stable-diffusion-stability-ai', '.', '/']
could you solve it? @simbrams
Has anyone figured out how to redeploy this repo?
@simbrams
I haven't dig further since last time. I had some errors that looked like corrupted model weights.
Next time, I will probably download the remote cog container files from replicate, copy paste all the files and try to run it again.
I am working on a download-weights.py file to make the whole deployment easier. The file is a work in progress but should fix some of the bugs mentioned here.
Hi @philz1337x! Congrats on your impressive work. I've been eager to try it out, but I've hit a snag when attempting to run the model. I've tried both the cod
method and the SD webui method, but I'm encountering issues with both.
cod method:
download-weights.py
gives me a RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check
.cog predict -I image
command, I get an AssertionError: Couldn't find Stable Diffusion in any of: ['/src/repositories/stable-diffusion-stability-ai', '.', '/']
. (I tried to put my SD folder into the current repo → doesn't work)SD webui method:
webui.sh
file in the current repo and attempted to run it instead of the one present on the SD repo, but it gives me a RuntimeError: Torch is not able to use GPU; add --skip-torch-cuda-test to COMMANDLINE_ARGS variable to disable this check
.I'm feeling a bit stuck here. Do you have any suggestions on how to proceed? Thanks!
Hey @philz1337x could you solve the redeploying issue? download_weights still does not work.
as of the current version of the repo, you need to add the vae to download weights:
# VAE
download_file(
"https://huggingface.co/stabilityai/sd-vae-ft-mse-original/resolve/main/vae-ft-mse-840000-ema-pruned.safetensors?download=true",
"models/VAE",
"vae-ft-mse-840000-ema-pruned.safetensors"
)
You also need to clone the multidiffusion upscaler and controlnet extensions to extensions/
.
For the controlnet extension, the most recent version does not work with this repo, so clone tag 1.1.436:
git clone --depth 1 --branch 1.1.436 git@github.com:Mikubill/sd-webui-controlnet.git
also make sure docker has access to your gpu. for me that's by using nvidia-docker
and running cog
with sudo
for fellow nix users:
virtualisation.docker = {
enable = true;
enableOnBoot = false;
enableNvidia = true;
rootless = {
enable = true;
setSocketVariable = false;
daemon.settings = {
runtimes = {
nvidia = {
path = "${pkgs.nvidia-docker}/bin/nvidia-container-runtime";
};
};
};
};
};
@e-cal added a great download-weights.py file to make all this easier!
=> CACHED [stage-1 6/11] RUN --mount=type=bind,from=deps,source=/dep,target=/dep cp -rf /dep/ $(pyenv prefix)/lib/python/site-packages || true 0.0s => CACHED [stage-1 7/11] RUN git config --global --add safe.directory /src 0.0s => CACHED [stage-1 8/11] RUN git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui /stable-diffusion-webui && cd /stable-diffusion-webui && git checkout 310d6b9075c6edb3b884bd2a41 0.0s => CACHED [stage-1 9/11] RUN git clone https://github.com/LLSean/cog-A1111-webui /cog-sd-webui 0.0s => CACHED [stage-1 10/11] RUN python /cog-sd-webui/init_env.py --skip-torch-cuda-test 0.0s => CACHED [stage-1 11/11] WORKDIR /src 0.0s => preparing layers for inline cache 0.0s => exporting to image 0.0s => => exporting layers 0.0s => => writing image sha256:6143abae3dbac3d3e24884eed7e8c5893db329a2ee5d06fd0964a654322d9e24 0.0s => => naming to docker.io/library/cog-clarity-upscaler-base 0.0s
Starting Docker image cog-clarity-upscaler-base and running setup()... Traceback (most recent call last): File "/root/.pyenv/versions/3.10.4/lib/python3.10/site-packages/cog/server/worker.py", line 189, in _setup run_setup(self._predictor) File "/root/.pyenv/versions/3.10.4/lib/python3.10/site-packages/cog/predictor.py", line 70, in run_setup predictor.setup() File "/src/predict.py", line 22, in setup initialize.imports() File "/src/modules/initialize.py", line 24, in imports from modules import paths, timer, import_hook, errors # noqa: F401 File "/src/modules/paths.py", line 34, in
assert sd_path is not None, f"Couldn't find Stable Diffusion in any of: {possible_sd_paths}"
AssertionError: Couldn't find Stable Diffusion in any of: ['/src/repositories/stable-diffusion-stability-ai', '.', '/']