279 lines
5.4 KiB
Markdown
279 lines
5.4 KiB
Markdown
# DreamOmni2 服务器部署指南
|
||
|
||
## 📋 服务器信息
|
||
- **IP**: 156.225.27.52
|
||
- **端口**: 5204
|
||
- **用户**: root
|
||
- **密码**: Jsjj@8888
|
||
|
||
## 🚀 快速部署步骤
|
||
|
||
### 步骤 1: SSH 连接到服务器
|
||
|
||
```bash
|
||
ssh -p 5204 root@156.225.27.52
|
||
```
|
||
|
||
输入密码: `Jsjj@8888`
|
||
|
||
### 步骤 2: 上传部署脚本
|
||
|
||
**方法一: 使用 SCP 上传**
|
||
|
||
在你的本地电脑终端执行：
|
||
|
||
```bash
|
||
scp -P 5204 dreamomni2-deploy.sh root@156.225.27.52:~/
|
||
```
|
||
|
||
**方法二: 直接在服务器创建**
|
||
|
||
SSH连接后，在服务器上执行：
|
||
|
||
```bash
|
||
cat > dreamomni2-deploy.sh << 'EOF'
|
||
# 这里粘贴 dreamomni2-deploy.sh 的完整内容
|
||
EOF
|
||
|
||
chmod +x dreamomni2-deploy.sh
|
||
```
|
||
|
||
### 步骤 3: 执行部署脚本
|
||
|
||
```bash
|
||
bash dreamomni2-deploy.sh
|
||
```
|
||
|
||
脚本会自动完成以下任务：
|
||
1. ✅ 检查系统信息和GPU
|
||
2. ✅ 安装系统依赖
|
||
3. ✅ 安装 Miniconda
|
||
4. ✅ 创建独立的 Python 虚拟环境
|
||
5. ✅ 安装 PyTorch (自动检测CUDA版本)
|
||
6. ✅ 克隆 DreamOmni2 仓库
|
||
7. ✅ 安装所有依赖包
|
||
8. ✅ 下载模型权重
|
||
|
||
## 📦 手动部署步骤 (如果脚本出现问题)
|
||
|
||
### 1. 连接服务器并更新系统
|
||
|
||
```bash
|
||
ssh -p 5204 root@156.225.27.52
|
||
apt update && apt upgrade -y
|
||
apt install -y git wget build-essential
|
||
```
|
||
|
||
### 2. 检查 GPU
|
||
|
||
```bash
|
||
nvidia-smi
|
||
```
|
||
|
||
### 3. 安装 Miniconda
|
||
|
||
```bash
|
||
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
|
||
bash Miniconda3-latest-Linux-x86_64.sh -b -p $HOME/miniconda3
|
||
export PATH="$HOME/miniconda3/bin:$PATH"
|
||
conda init bash
|
||
source ~/.bashrc
|
||
```
|
||
|
||
### 4. 创建虚拟环境
|
||
|
||
```bash
|
||
conda create -n dreamomni2 python=3.10 -y
|
||
conda activate dreamomni2
|
||
```
|
||
|
||
### 5. 安装 PyTorch (根据CUDA版本)
|
||
|
||
**CUDA 12.x:**
|
||
```bash
|
||
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
|
||
```
|
||
|
||
**CUDA 11.8:**
|
||
```bash
|
||
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
|
||
```
|
||
|
||
### 6. 克隆 DreamOmni2
|
||
|
||
```bash
|
||
cd ~
|
||
git clone https://github.com/dvlab-research/DreamOmni2.git
|
||
cd DreamOmni2
|
||
```
|
||
|
||
### 7. 安装依赖
|
||
|
||
```bash
|
||
pip install -r requirements.txt
|
||
pip install transformers diffusers accelerate safetensors huggingface_hub
|
||
```
|
||
|
||
### 8. 下载模型
|
||
|
||
```bash
|
||
mkdir -p models
|
||
cd models
|
||
pip install -U huggingface_hub
|
||
huggingface-cli download xiabs/DreamOmni2 --local-dir ./dreamomni2-model
|
||
```
|
||
|
||
## 🎯 使用 DreamOmni2
|
||
|
||
### 激活环境
|
||
|
||
```bash
|
||
conda activate dreamomni2
|
||
cd ~/DreamOmni2
|
||
```
|
||
|
||
### 运行推理示例
|
||
|
||
```bash
|
||
python inference.py \
|
||
--source_image ./examples/source.jpg \
|
||
--reference_image ./examples/reference.jpg \
|
||
--prompt "将源图像的风格改为参考图像的风格" \
|
||
--output ./output.jpg
|
||
```
|
||
|
||
### 创建简单测试脚本
|
||
|
||
```python
|
||
# test_dreamomni2.py
|
||
import torch
|
||
from PIL import Image
|
||
|
||
def test_model():
|
||
print("正在测试 DreamOmni2...")
|
||
print(f"PyTorch 版本: {torch.__version__}")
|
||
print(f"CUDA 可用: {torch.cuda.is_available()}")
|
||
if torch.cuda.is_available():
|
||
print(f"GPU 数量: {torch.cuda.device_count()}")
|
||
print(f"GPU 名称: {torch.cuda.get_device_name(0)}")
|
||
|
||
# 这里添加 DreamOmni2 的实际推理代码
|
||
print("✅ 环境测试通过！")
|
||
|
||
if __name__ == "__main__":
|
||
test_model()
|
||
```
|
||
|
||
运行测试：
|
||
```bash
|
||
python test_dreamomni2.py
|
||
```
|
||
|
||
## 🔧 常见问题
|
||
|
||
### Q1: 如何检查部署状态？
|
||
|
||
```bash
|
||
conda activate dreamomni2
|
||
python -c "import torch; print(f'PyTorch: {torch.__version__}, CUDA: {torch.cuda.is_available()}')"
|
||
```
|
||
|
||
### Q2: 如何查看显存使用情况？
|
||
|
||
```bash
|
||
watch -n 1 nvidia-smi
|
||
```
|
||
|
||
### Q3: 模型下载太慢怎么办？
|
||
|
||
可以使用国内镜像或者手动下载：
|
||
|
||
```bash
|
||
export HF_ENDPOINT=https://hf-mirror.com
|
||
huggingface-cli download xiabs/DreamOmni2 --local-dir ./models/dreamomni2-model
|
||
```
|
||
|
||
### Q4: 如何配置 API 服务？
|
||
|
||
创建 API 服务脚本：
|
||
|
||
```python
|
||
# api_server.py
|
||
from fastapi import FastAPI, File, UploadFile
|
||
from fastapi.responses import FileResponse
|
||
import uvicorn
|
||
|
||
app = FastAPI()
|
||
|
||
@app.post("/generate")
|
||
async def generate_image(
|
||
source: UploadFile = File(...),
|
||
reference: UploadFile = File(...),
|
||
prompt: str = ""
|
||
):
|
||
# 这里添加 DreamOmni2 推理逻辑
|
||
return {"status": "success"}
|
||
|
||
if __name__ == "__main__":
|
||
uvicorn.run(app, host="0.0.0.0", port=8000)
|
||
```
|
||
|
||
安装依赖并运行：
|
||
```bash
|
||
pip install fastapi uvicorn python-multipart
|
||
python api_server.py
|
||
```
|
||
|
||
## 💾 硬件要求
|
||
|
||
- **推荐**: NVIDIA RTX 3090 / 4090 或更高
|
||
- **最低显存**: 16GB VRAM
|
||
- **系统内存**: 32GB+ RAM
|
||
- **存储空间**: 50GB+ (模型 + 依赖)
|
||
|
||
## 🔐 安全建议
|
||
|
||
1. **修改SSH密码**:
|
||
```bash
|
||
passwd
|
||
```
|
||
|
||
2. **配置防火墙**:
|
||
```bash
|
||
ufw allow 5204
|
||
ufw enable
|
||
```
|
||
|
||
3. **禁用root远程登录** (可选，但推荐):
|
||
```bash
|
||
# 创建新用户
|
||
adduser aiuser
|
||
usermod -aG sudo aiuser
|
||
# 修改 /etc/ssh/sshd_config
|
||
# PermitRootLogin no
|
||
```
|
||
|
||
## 📞 获取帮助
|
||
|
||
- **GitHub Issues**: https://github.com/dvlab-research/DreamOmni2/issues
|
||
- **Hugging Face**: https://huggingface.co/xiabs/DreamOmni2
|
||
- **官网**: https://www.dreamomni2.com
|
||
|
||
## 🎉 完成后验证
|
||
|
||
```bash
|
||
# 激活环境
|
||
conda activate dreamomni2
|
||
|
||
# 检查安装
|
||
cd ~/DreamOmni2
|
||
python -c "import torch; print('✅ PyTorch安装成功')"
|
||
python -c "from transformers import AutoModel; print('✅ Transformers安装成功')"
|
||
python -c "from diffusers import StableDiffusionPipeline; print('✅ Diffusers安装成功')"
|
||
|
||
# 查看模型文件
|
||
ls -lh ~/DreamOmni2/models/dreamomni2-model/
|
||
```
|
||
|
||
如果所有检查都通过，说明部署成功！🎊
|