stable-diffusion-webui 加载模型文件

背景

stable-diffusion-webui 安装完毕后，默认的模型生成的效果图并不理想，可以根据具体需求加载指定的模型文件。国内 modelscope 下载速度较快，以该站为例进行介绍

操作步骤

1. 找到指定的模型文件
   在 https://modelscope.cn/models 中查找文本生成图片标签的模型，根据自己喜好点进模型详情页，初次使用，我们可以挑选一些模型文件较小的文件用于测试
   
2. 找到模型文件的下载地址
   找到模型文件列表中的.safetensors文件，例如：flux1-dev.safetensors，右键单击下载按钮，复制链接地址。
3. 下载该模型文件
   进入服务器路径/stable-diffusion-webui-docker/data/models/Stable-diffusion下，wget https://modelscope.cn/models/black-forest-labs/FLUX.1-dev/resolve/master/flux1-dev.safetensors 即可
4. 重启 stable-diffusion-webui
   在左上角的模型下拉列表框中选中切换即可

下载 Lora 模型

modelscope 上模型标签有 LoRA 的都是支持 Lora 的模型，进入模型详情页，通过上一节的方法找到 Lora 模型的下载链接，进入服务器路径/stable-diffusion-webui-docker/data/models/Lora下，wget 下载即可。webui页面上，单击 Lora 页签，单击刷新

使用时，单击模型即可

运行时报错

Expected all tensors to be on the same device, but found at least two devices

auto-1  |       File "/stable-diffusion-webui/modules/sd_hijack.py", line 348, in forward
auto-1  |         inputs_embeds = self.wrapped(input_ids)
auto-1  |       File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1532, in _wrapped_call_impl
auto-1  |         return self._call_impl(*args, **kwargs)
auto-1  |       File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1541, in _call_impl
auto-1  |         return forward_call(*args, **kwargs)
auto-1  |       File "/opt/conda/lib/python3.10/site-packages/torch/nn/modules/sparse.py", line 163, in forward
auto-1  |         return F.embedding(
auto-1  |       File "/opt/conda/lib/python3.10/site-packages/torch/nn/functional.py", line 2264, in embedding
auto-1  |         return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
auto-1  |     RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument index in method wrapper_CUDA__index_select)

参考 https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/16263 解决，我的解决方案是修改 /stable-diffusion-webui-docker/docker-compose.yml，指定 --device-id=0，如下：

services:
  download:
    build: ./services/download/
    profiles: ["download"]
    volumes:
      - *v1

  auto: &automatic
    <<: *base_service
    profiles: ["auto"]
    build: ./services/AUTOMATIC1111
    image: sd-auto:78
    environment:
      - CLI_ARGS=--allow-code --medvram --xformers --enable-insecure-extension-access --api --device-id 0

safetensors_rust.SafetensorError: Error while deserializing header: MetadataIncompleteBuffer

auto-1  |   warnings.warn(
auto-1  | creating model quickly: SafetensorError
auto-1  | Traceback (most recent call last):
auto-1  |   File "/opt/conda/lib/python3.10/threading.py", line 973, in _bootstrap
auto-1  |     self._bootstrap_inner()
auto-1  |   File "/opt/conda/lib/python3.10/threading.py", line 1016, in _bootstrap_inner
auto-1  |     self.run()
auto-1  |   File "/opt/conda/lib/python3.10/threading.py", line 953, in run
auto-1  |     self._target(*self._args, **self._kwargs)
auto-1  |   File "/stable-diffusion-webui/modules/initialize.py", line 149, in load_model
auto-1  |     shared.sd_model  # noqa: B018
auto-1  |   File "/stable-diffusion-webui/modules/shared_items.py", line 175, in sd_model
auto-1  |     return modules.sd_models.model_data.get_sd_model()
auto-1  |   File "/stable-diffusion-webui/modules/sd_models.py", line 620, in get_sd_model
auto-1  |     load_model()
auto-1  |   File "/stable-diffusion-webui/modules/sd_models.py", line 723, in load_model
auto-1  |     sd_model = instantiate_from_config(sd_config.model)
auto-1  |   File "/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/util.py", line 89, in instantiate_from_config
auto-1  |     return get_obj_from_str(config["target"])(**config.get("params", dict()))
auto-1  |   File "/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/models/diffusion/ddpm.py", line 563, in __init__
auto-1  |     self.instantiate_cond_stage(cond_stage_config)
auto-1  |   File "/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/models/diffusion/ddpm.py", line 630, in instantiate_cond_stage
auto-1  |     model = instantiate_from_config(config)
auto-1  |   File "/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/util.py", line 89, in instantiate_from_config
auto-1  |     return get_obj_from_str(config["target"])(**config.get("params", dict()))
auto-1  |   File "/stable-diffusion-webui/repositories/stable-diffusion-stability-ai/ldm/modules/encoders/modules.py", line 104, in __init__
auto-1  |     self.transformer = CLIPTextModel.from_pretrained(version)
auto-1  |   File "/opt/conda/lib/python3.10/site-packages/transformers/modeling_utils.py", line 2604, in from_pretrained
auto-1  |     state_dict = load_state_dict(resolved_archive_file)
auto-1  |   File "/opt/conda/lib/python3.10/site-packages/transformers/modeling_utils.py", line 450, in load_state_dict
auto-1  |     with safe_open(checkpoint_file, framework="pt") as f:
auto-1  | safetensors_rust.SafetensorError: Error while deserializing header: MetadataIncompleteBuffer

参考
https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/14267
https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/15568
大概有两种原因，要么是 safetensors 模型文件损坏，要么是文件 SHA 值校验不正确，有个暴力优化的办法是，直接删除 /stable-diffusion-webui-docker/data/.cache 隐藏目录，然后重启