你开心就好
Jetson-Orin-Nano-Super(7) stable-diffusion-webui 部署
编辑
2025-09-09
记录知识
0

目录

jetson ai lab
实战
参考
总结

上一个文章参考了网络的文章实战部署了qwen3 8b的llm模型，进一步的本文继续参考网络资源，上手在orin上部署stable diffusion

jetson ai lab

nvidia有一个ai lab，可以游玩很多的ai相关内容

https://www.jetson-ai-lab.com/

本文主要实战部署stable-diffusion，所以参考文档如下

https://www.jetson-ai-lab.com/tutorial_stable-diffusion.html

可以看到，这里面会使用jetson containers的git仓库，先进去浏览一番

https://github.com/dusty-nv/jetson-containers.git

可以看到这个仓库有很多的ai示例可以尝试

image.png

这里找到stable diffusion的示例，其实际上是拉取了docker hub的容器镜像，如下

https://hub.docker.com/r/dustynv/stable-diffusion-webui

可以看到当前最新的版本是r36.4.0

下面开始根据文档的步骤演示运行stable-diffusion-webui

实战

首先国内的docker需要设置一下proxy。

# mkdir -p /etc/systemd/system/docker.service.d
# cat /etc/systemd/system/docker.service.d/http-proxy.conf
[Service]
Environment="HTTP_PROXY=http://127.0.0.1:9981"
Environment="HTTPS_PROXY=http://127.0.0.1:9981"
# systemctl daemon-reload
# systemctl restart docker

这里不同的proxy换成不同端口即可。

然后按照文档，这里尝试安装最新的版本r36.4.0，如下

# jetson-containers run dustynv/stable-diffusion-webui:r36.4.0

但是发现此版本的python的numpy要求，venv无法提供，所以降级版本到r36.2.0，如下

# jetson-containers run dustynv/stable-diffusion-webui:r36.2.0

因为在国内，所以默认模型下载不了，这是安装上述配置挂上代理后下载的。

下载完成之后，运行后发现docker的运行参数是如下

docker run --runtime nvidia --env NVIDIA_DRIVER_CAPABILITIES=compute,utility,graphics -it --rm --network host --shm-size=8g --volume /root/jetson-containers/jetson-containers/packages/diffusion/stable-diffusion-webui/openai/:/opt/stable-diffusion-webui/openai/ --volume /tmp/argus_socket:/tmp/argus_socket --volume /etc/enctune.conf:/etc/enctune.conf --volume /etc/nv_tegra_release:/etc/nv_tegra_release --volume /tmp/nv_jetson_model:/tmp/nv_jetson_model --volume /var/run/dbus:/var/run/dbus --volume /var/run/avahi-daemon/socket:/var/run/avahi-daemon/socket --volume /var/run/docker.sock:/var/run/docker.sock --volume /root/jetson-containers/jetson-containers/data:/data -v /etc/localtime:/etc/localtime:ro -v /etc/timezone:/etc/timezone:ro --device /dev/snd --device /dev/bus/usb --device /dev/i2c-0 --device /dev/i2c-1 --device /dev/i2c-2 --device /dev/i2c-4 --device /dev/i2c-5 --device /dev/i2c-7 --device /dev/i2c-9 -v /run/jtop.sock:/run/jtop.sock --name jetson_container_20250910_222642 dustynv/stable-diffusion-webui:r36.2.0

第一个问题出现了，容器内的模型无法下载，这里观察到默认的data在/root/jetson-containers/jetson-containers/data上，如下

--volume /root/jetson-containers/jetson-containers/data:/data

所以进入目录手动下载模型到指定目录，如下

# cd /root/jetson-containers/jetson-containers/data/models/stable-diffusion/models/Stable-diffusion
# wget https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned-emaonly.safetensors

下载后继续尝试运行，发现CLIPTokenizer无法加载，日志如下

Python 3.10.12 (main, Jun 11 2023, 05:26:28) [GCC 11.4.0]
Version: v1.7.0
Commit hash: cf2772fab0af5573da775e7437e6acdca424f26e
Launching Web UI with arguments: --data=/data/models/stable-diffusion --enable-insecure-extension-access --xformers --listen --port=7860
Style database not found: /data/models/stable-diffusion/styles.csv
Loading weights [6ce0161689] from /data/models/stable-diffusion/models/Stable-diffusion/v1-5-pruned-emaonly.safetensors
/opt/stable-diffusion-webui/extensions-builtin/stable-diffusion-webui-tensorrt/ui_trt.py:64: GradioDeprecationWarning: The `style` method is deprecated. Please set these arguments in the constructor instead.
  with gr.Row().style(equal_height=False):
Running on local URL:  http://0.0.0.0:7860

To create a public link, set `share=True` in `launch()`.
Startup time: 15.5s (prepare environment: 3.1s, import torch: 4.8s, import gradio: 1.9s, setup paths: 1.5s, initialize shared: 0.2s, other imports: 1.3s, setup codeformer: 0.1s, load scripts: 0.9s, create ui: 0.9s, gradio launch: 0.8s).
Creating model from config: /opt/stable-diffusion-webui/configs/v1-inference.yaml
/opt/stable-diffusion-webui/openai/clip-vit-large-patch14
creating model quickly: OSError
Traceback (most recent call last):
  File "/usr/lib/python3.10/threading.py", line 973, in _bootstrap
    self._bootstrap_inner()
  File "/usr/lib/python3.10/threading.py", line 1016, in _bootstrap_inner
    self.run()
  File "/usr/lib/python3.10/threading.py", line 953, in run
    self._target(*self._args, **self._kwargs)
  File "/opt/stable-diffusion-webui/modules/initialize.py", line 147, in load_model
    shared.sd_model  # noqa: B018
  File "/opt/stable-diffusion-webui/modules/shared_items.py", line 128, in sd_model
    return modules.sd_models.model_data.get_sd_model()
  File "/opt/stable-diffusion-webui/modules/sd_models.py", line 531, in get_sd_model
    load_model()
  File "/opt/stable-diffusion-webui/modules/sd_models.py", line 634, in load_model
    sd_model = instantiate_from_config(sd_config.model)
  File "/opt/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/util.py", line 89, in instantiate_from_config
    return get_obj_from_str(config["target"])(**config.get("params", dict()))
  File "/opt/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/models/diffusion/ddpm.py", line 563, in __init__
    self.instantiate_cond_stage(cond_stage_config)
  File "/opt/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/models/diffusion/ddpm.py", line 630, in instantiate_cond_stage
    model = instantiate_from_config(config)
  File "/opt/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/util.py", line 89, in instantiate_from_config
    return get_obj_from_str(config["target"])(**config.get("params", dict()))
  File "/opt/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/modules/encoders/modules.py", line 103, in __init__
    self.tokenizer = CLIPTokenizer.from_pretrained(version)
  File "/usr/local/lib/python3.10/dist-packages/transformers/tokenization_utils_base.py", line 1809, in from_pretrained
    raise EnvironmentError(
OSError: Can't load tokenizer for 'openai/clip-vit-large-patch14'. If you were trying to load it from 'https://huggingface.co/models', make sure you don't have a local directory with the same name. Otherwise, make sure 'openai/clip-vit-large-patch14' is the correct path to a directory containing all relevant files for a CLIPTokenizer tokenizer.

根据日志初步怀疑也是docker内的环境拿到openai/clip-vit-large-patch14，所以也是手动下载

# cd /root/jetson-containers/jetson-containers/packages/diffusion/stable-diffusion-webui
# mkdir openai && cd openai
# apt install git-lfs
# git clone https://huggingface.co/openai/clip-vit-large-patch14

经过漫长的等待，终于拉取了其13G的模型内容，如下

# cd /root/jetson-containers/jetson-containers/packages/diffusion/stable-diffusion-webui/openai/clip-vit-large-patch14    .

到这里还需要判断和分享一下日志报错的情况

这里根据报错的日志，看到launch.py的运行参数是如下

Launching Web UI with arguments: --data=/data/models/stable-diffusion --enable-insecure-extension-access --xformers --listen --port=7860

所以其实际运行的命令可能是如下

# python3 launch.py --data=/data/models/stable-diffusion --enable-insecure-extension-access --xformers --listen --port=7860

根据python的堆栈可以知道代码出错在/usr/local/lib/python3.10/dist-packages/transformers/tokenization_utils_base.py的1809行，进入确认一下，找到from_pretrained函数的声明，如下

def from_pretrained(cls, pretrained_model_name_or_path: Union[str, os.PathLike], *init_inputs, **kwargs):

可以看到，这是pretrained_model_name_or_path: Union[str, os.PathLike]计算的路径，而pretrained_model_name_or_path可以从如下确认

  File "/opt/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/modules/encoders/modules.py", line 103, in __init__
    self.tokenizer = CLIPTokenizer.from_pretrained(version)

继续查看modules.py，确认version的值

def __init__(self, version="openai/clip-vit-large-patch14", device="cuda", max_length=77,
                 freeze=True, layer="last", layer_idx=None):  # clip-vit-base-patch32

合理猜测需要的路径实际上应该是容器里这个位置

/opt/stable-diffusion-webui/openai/clip-vit-large-patch14

为了保险，这里打上python的print再确认一遍，这里在from_pretrained打印pretrained_model_name_or_path的值，如下

print(os.path.abspath(pretrained_model_name_or_path))

手动运行，得到print打印，如下

# python3 launch.py --data=/data/models/stable-diffusion --enable-insecure-extension-access --xformers --listen --port=7860
/opt/stable-diffusion-webui/openai/clip-vit-large-patch14

猜测完全正确，需要在/opt/stable-diffusion-webui有对应的目录，现在返回去看docker run的参数，可以发现，/opt/stable-diffusion-webui并不是对外挂载的，所以需要新增一条volume参数，将之前下载的模型位置挂载到docker的这个目录上，如下

--volume /root/jetson-containers/jetson-containers/packages/diffusion/stable-diffusion-webui/openai/:/opt/stable-diffusion-webui/openai/

所以总的运行命令如下

docker run --runtime nvidia --env NVIDIA_DRIVER_CAPABILITIES=compute,utility,graphics -it --rm --network host --shm-size=8g --volume /root/jetson-containers/jetson-containers/packages/diffusion/stable-diffusion-webui/openai/:/opt/stable-diffusion-webui/openai/ --volume /tmp/argus_socket:/tmp/argus_socket --volume /etc/enctune.conf:/etc/enctune.conf --volume /etc/nv_tegra_release:/etc/nv_tegra_release --volume /tmp/nv_jetson_model:/tmp/nv_jetson_model --volume /var/run/dbus:/var/run/dbus --volume /var/run/avahi-daemon/socket:/var/run/avahi-daemon/socket --volume /var/run/docker.sock:/var/run/docker.sock --volume /root/jetson-containers/jetson-containers/data:/data -v /etc/localtime:/etc/localtime:ro -v /etc/timezone:/etc/timezone:ro --device /dev/snd --device /dev/bus/usb --device /dev/i2c-0 --device /dev/i2c-1 --device /dev/i2c-2 --device /dev/i2c-4 --device /dev/i2c-5 --device /dev/i2c-7 --device /dev/i2c-9 -v /run/jtop.sock:/run/jtop.sock --name jetson_container_20250910_222642 dustynv/stable-diffusion-webui:r36.2.0

此时运行日志如下

Python 3.10.12 (main, Jun 11 2023, 05:26:28) [GCC 11.4.0]
Version: v1.7.0
Commit hash: cf2772fab0af5573da775e7437e6acdca424f26e
Launching Web UI with arguments: --data=/data/models/stable-diffusion --enable-insecure-extension-access --xformers --listen --port=7860
Style database not found: /data/models/stable-diffusion/styles.csv
Loading weights [6ce0161689] from /data/models/stable-diffusion/models/Stable-diffusion/v1-5-pruned-emaonly.safetensors
/opt/stable-diffusion-webui/extensions-builtin/stable-diffusion-webui-tensorrt/ui_trt.py:64: GradioDeprecationWarning: The `style` method is deprecated. Please set these arguments in the constructor instead.
  with gr.Row().style(equal_height=False):
Running on local URL:  http://0.0.0.0:7860

To create a public link, set `share=True` in `launch()`.
Startup time: 18.2s (prepare environment: 3.4s, import torch: 5.6s, import gradio: 2.2s, setup paths: 1.7s, initialize shared: 0.3s, other imports: 1.8s, setup codeformer: 0.2s, load scripts: 1.1s, create ui: 0.9s, gradio launch: 0.8s).
Creating model from config: /opt/stable-diffusion-webui/configs/v1-inference.yaml
Applying attention optimization: xformers... done.
Model loaded in 10.8s (load weights from disk: 2.2s, create model: 0.8s, apply weights to model: 6.5s, apply half(): 0.2s, load textual inversion embeddings: 0.6s, calculate empty prompt: 0.3s).

没有看到报错了，同时可以看到默认开放端口在0.0.0.0:7860，所以可以同一网络环境下打开这个页面，做一下尝试.

这里简单生成了一下麒麟嵌入式系统的图片，还是比较准确的，如下。

image.png

再生成一张，查看一下gpu占用情况

image.png

image.png

可以看到，内存5个多G，GPU全跑，整体比qwen3 8b要顺利很多，orin nano运行stable diffusion 并不是很吃力。

最后，其实jetson-ai-lab的文档也告诉你了，容器是这样如下运行的，

cd /opt/stable-diffusion-webui && python3 launch.py \
  --data=/data/models/stable-diffusion \
  --enable-insecure-extension-access \
  --xformers \
  --listen \
  --port=7860

参考

https://www.jetson-ai-lab.com/tutorial_stable-diffusion.html
https://zhuanlan.zhihu.com/p/1889443308834624504

总结

本次参考网络资源实战了stable diffusion，相比于ollma，因为是容器运行，所以代理的问题比较难搞，但是总归来说有办法的，整体还是非常轻松能够跑起来的。使用两个小时下来，orin 发热不严重，完全满足可以长期运行。