Skip to content

Commit 2d87685

Browse files
xzuoqixzuoqi
authored
Merge pull request #1213 from EESN-W/deploy-error
fix: newcodeblock format error
2 parents 1ea314c + 21d51cd commit 2d87685

File tree

1 file changed

+115
-102
lines changed

1 file changed

+115
-102
lines changed

‎docs/common/ai/_stable_diffusion_convert.mdx‎

Lines changed: 115 additions & 102 deletions
Original file line numberDiff line numberDiff line change
@@ -17,32 +17,35 @@ Stable Diffusion 是一种基于潜在扩散模型的文本到图像生成模型
1717
- **使用modelscope下载模型文件**
1818

1919
- **创建存储模型文件的目录**
20-
<NewCodeBlock tip="Linux PC" type="PC">
2120

22-
```bash
23-
mkdir sd-lcm-rknn && cd sd-lcm-rknn
24-
```
21+
<NewCodeBlock tip="Linux PC" type="PC">
2522

26-
</NewCodeBlock>
23+
```bash
24+
mkdir sd-lcm-rknn && cd sd-lcm-rknn
25+
```
26+
27+
</NewCodeBlock>
2728

28-
- **pip安装modelscope包**
29-
<NewCodeBlock tip="Linux PC" type="PC">
29+
- **pip安装modelscope包**
3030

31-
```bash
32-
# 尽量使用较新的python版本避免兼容问题
33-
pip3 install modelscope
34-
```
31+
<NewCodeBlock tip="Linux PC" type="PC">
3532

36-
</NewCodeBlock>
33+
```bash
34+
# 尽量使用较新的python版本避免兼容问题
35+
pip3 install modelscope
36+
```
3737

38-
- **克隆Stable-Diffusion-LCM_RKNN仓库**
39-
<NewCodeBlock tip="Linux PC" type="PC">
38+
</NewCodeBlock>
4039

41-
```bash
42-
modelscope download --model radxa/Stable-Diffusion-LCM_RKNN
43-
```
40+
- **克隆Stable-Diffusion-LCM_RKNN仓库**
4441

45-
</NewCodeBlock>
42+
<NewCodeBlock tip="Linux PC" type="PC">
43+
44+
```bash
45+
modelscope download --model radxa/Stable-Diffusion-LCM_RKNN
46+
```
47+
48+
</NewCodeBlock>
4649

4750
## （可选）模型转换
4851

@@ -51,122 +54,132 @@ Stable Diffusion 是一种基于潜在扩散模型的文本到图像生成模型
5154
- **从HuggingFace下载ONNX模型并转换为RKNN模型**
5255

5356
- **创建存储模型文件的目录**
54-
<NewCodeBlock tip="Linux PC" type="PC">
5557

56-
```bash
57-
mkdir sd-lcm-rknn && cd sd-lcm-rknn
58-
```
58+
<NewCodeBlock tip="Linux PC" type="PC">
5959

60-
</NewCodeBlock>
60+
```bash
61+
mkdir sd-lcm-rknn && cd sd-lcm-rknn
62+
```
6163

62-
- **克隆模型仓库**
63-
<NewCodeBlock tip="Linux PC" type="PC">
64+
</NewCodeBlock>
6465

65-
```bash
66-
# 需要使用git lfs，如未安装请自行安装
67-
git lfs install
68-
git clone https://huggingface.co/thanhtantran/Stable-Diffusion-1.5-LCM-ONNX-RKNN2
69-
```
66+
- **克隆模型仓库**
7067

71-
</NewCodeBlock>
68+
<NewCodeBlock tip="Linux PC" type="PC">
7269

73-
- **激活虚拟环境**
74-
<NewCodeBlock tip="Linux PC" type="PC">
70+
```bash
71+
# 需要使用git lfs，如未安装请自行安装
72+
git lfs install
73+
git clone https://huggingface.co/thanhtantran/Stable-Diffusion-1.5-LCM-ONNX-RKNN2
74+
```
7575

76-
```bash
77-
conda activate your_rknn_env
78-
```
76+
</NewCodeBlock>
7977

80-
</NewCodeBlock>
78+
- **激活虚拟环境**
8179

82-
- **可先运行 _run_onnx-lcm.py_ 测试模型完整性**
83-
<NewCodeBlock tip="Linux PC" type="PC">
80+
<NewCodeBlock tip="Linux PC" type="PC">
8481

85-
```bash
82+
```bash
83+
conda activate your_rknn_env
84+
```
85+
86+
</NewCodeBlock>
87+
88+
- **可先运行 _run_onnx-lcm.py_ 测试模型完整性**
89+
90+
<NewCodeBlock tip="Linux PC" type="PC">
91+
92+
```bash
8693
# 使用-h参数查看参数帮助
8794
python run_onnx-lcm.py -i ./model -o ./images --prompt "Majestic mountain landscape with snow-capped peaks, autumn foliage in vibrant reds and oranges, a turquoise river winding through a valley, crisp and serene atmosphere, ultra-realistic style."
88-
```
89-
90-
</NewCodeBlock>
91-
92-
- **运行 _convert-onnx-to-rknn.py_ 转换模型**
93-
<NewCodeBlock tip="Linux PC" type="PC">
94-
95-
```bash
96-
# 使用-h参数查看参数帮助，将N替换为你实际需要的分辨率，转换之后模型只可输出该分辨率图像
97-
python convert-onnx-to-rknn.py -i ./model -r NxN
98-
```
99-
100-
</NewCodeBlock>
101-
102-
- **将文件整理为下面的目录结构，即可进行下一环节**
103-
104-
```txt
105-
---sd-lcm-rknn
106-
---model
107-
---scheduler
108-
---scheduler_config.json
109-
---text_encoder
110-
---config.json
111-
---model.rknn
112-
---unet
113-
---config.json
114-
---model.rknn
115-
---vae_decoder
116-
---config.json
117-
---model.rknn
118-
---run_rknn-lcm.py
119-
```
95+
```
96+
97+
</NewCodeBlock>
98+
99+
- **运行 _convert-onnx-to-rknn.py_ 转换模型**
100+
101+
<NewCodeBlock tip="Linux PC" type="PC">
102+
103+
```bash
104+
# 使用-h参数查看参数帮助，将N替换为你实际需要的分辨率，转换之后模型只可输出该分辨率图像
105+
python convert-onnx-to-rknn.py -i ./model -r NxN
106+
```
107+
108+
</NewCodeBlock>
109+
110+
- **将文件整理为下面的目录结构，即可进行下一环节**
111+
112+
```txt
113+
---sd-lcm-rknn
114+
---model
115+
---scheduler
116+
---scheduler_config.json
117+
---text_encoder
118+
---config.json
119+
---model.rknn
120+
---unet
121+
---config.json
122+
---model.rknn
123+
---vae_decoder
124+
---config.json
125+
---model.rknn
126+
---run_rknn-lcm.py
127+
```
120128

121129
## 板端部署
122130

123131
- **将转换后的RKNN模型及可执行文件拷贝到板端**
124132

125133
- **进入板端对应目录**
126-
<NewCodeBlock tip="Radxa SBC" type="device">
127134

128-
```bash
129-
cd sd-lcm-rknn
130-
```
135+
<NewCodeBlock tip="Radxa SBC" type="device">
131136

132-
</NewCodeBlock>
137+
```bash
138+
cd sd-lcm-rknn
139+
```
133140

134-
- **创建python虚拟环境**
135-
<NewCodeBlock tip="Radxa SBC" type="device">
141+
</NewCodeBlock>
136142

137-
```bash
138-
python -m venv .venv
139-
```
143+
- **创建python虚拟环境**
140144

141-
</NewCodeBlock>
145+
<NewCodeBlock tip="Radxa SBC" type="device">
142146

143-
- **激活虚拟环境**
144-
<NewCodeBlock tip="Radxa SBC" type="device">
147+
```bash
148+
python -m venv .venv
149+
```
145150

146-
```bash
147-
source .venv/bin/activate
148-
```
151+
</NewCodeBlock>
149152

150-
</NewCodeBlock>
153+
- **激活虚拟环境**
151154

152-
- **安装相关依赖**
153-
<NewCodeBlock tip="Radxa SBC" type="device">
155+
<NewCodeBlock tip="Radxa SBC" type="device">
154156

155-
```bash
156-
pip3 install diffusers pillow "numpy<2.0" torch transformers rknn-toolkit-lite2
157-
```
157+
```bash
158+
source .venv/bin/activate
159+
```
160+
161+
</NewCodeBlock>
162+
163+
- **安装相关依赖**
164+
165+
<NewCodeBlock tip="Radxa SBC" type="device">
158166

159-
</NewCodeBlock>
167+
```bash
168+
pip3 install diffusers pillow "numpy<2.0" torch transformers rknn-toolkit-lite2
169+
```
170+
171+
</NewCodeBlock>
160172

161-
- **执行脚本**
162-
<NewCodeBlock tip="Radxa SBC" type="device">
173+
- **执行脚本**
163174

164-
```bash
165-
# 使用-h参数查看参数帮助，自行转换的模型需要替换分辨率参数
166-
python ./run_rknn-lcm.py -i ./model -o ./images -s 256x256 --prompt "Majestic mountain landscape with snow-capped peaks, autumn foliage in vibrant reds and oranges, a turquoise river winding through a valley, crisp and serene atmosphere, ultra-realistic style."
167-
```
175+
<NewCodeBlock tip="Radxa SBC" type="device">
176+
177+
```bash
178+
# 使用-h参数查看参数帮助，自行转换的模型需要替换分辨率参数
179+
python ./run_rknn-lcm.py -i ./model -o ./images -s 256x256 --prompt "Majestic mountain landscape with snow-capped peaks, autumn foliage in vibrant reds and oranges, a turquoise river winding through a valley, crisp and serene atmosphere, ultra-realistic style."
180+
```
168181

169-
</NewCodeBlock>
182+
</NewCodeBlock>
170183

171184
## 模型结果及性能分析
172185

0 commit comments

Comments
 (0)