30：【uv + torch】2026最快安装GPU PyTorch & 避开conda坑

# conda安装问题示例
# 安装速度慢
Collecting package metadata (current_repodata.json): done
Solving environment: failed with initial frozen solve. Retrying with flexible solve.
Solving environment: failed with repodata from current_repodata.json, will retry with next repodata source.

# 依赖冲突
Found conflicts! Looking for incompatible packages.
This can take several minutes.  Press CTRL-C to abort.

# 环境污染
CondaHTTPError: HTTP 000 CONNECTION FAILED for url <https://repo.anaconda.com/pkgs/main/win-64/current_repodata.json>

# uv安装速度示例
# 安装PyTorch GPU版仅需15秒左右
Installed 12 packages in 14.8s

# 安装uv（如果未安装）
pip install uv

# 创建并激活虚拟环境
uv venv
uv activate

# 安装PyTorch GPU版（CUDA 13.1）
uv pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu131

# 验证安装
python -c "import torch; print('PyTorch版本:', torch.__version__); print('CUDA可用:', torch.cuda.is_available())"

# 导出conda环境依赖
conda env export --no-builds > environment.yml

# 安装uv
pip install uv

# 创建uv虚拟环境
uv venv
uv activate

# 安装依赖
uv pip install -r requirements.txt

# 安装PyTorch GPU版
uv pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu131

#!/usr/bin/env python3
"""
uv性能优化工具
"""
import subprocess
import sys
import time

def benchmark_uv_vs_conda():
    """基准测试uv与conda的安装速度"""
    print("=== uv vs conda 安装速度基准测试 ===")
    
    # 测试包列表
    packages = ["numpy", "pandas", "matplotlib", "scikit-learn", "jupyter"]
    
    # 测试uv安装速度
    print("\n1. 测试uv安装速度...")
    start_time = time.time()
    result = subprocess.run(
        [sys.executable, "-m", "uv", "pip", "install"] + packages,
        capture_output=True,
        text=True
    )
    uv_time = time.time() - start_time
    print(f"uv安装耗时: {uv_time:.2f}秒")
    
    # 测试pip安装速度
    print("\n2. 测试pip安装速度...")
    start_time = time.time()
    result = subprocess.run(
        [sys.executable, "-m", "pip", "install"] + packages,
        capture_output=True,
        text=True
    )
    pip_time = time.time() - start_time
    print(f"pip安装耗时: {pip_time:.2f}秒")
    
    # 计算速度提升
    speedup = pip_time / uv_time if uv_time > 0 else 0
    print(f"\nuv相比pip速度提升: {speedup:.2f}倍")

def optimize_uv_config():
    """优化uv配置"""
    print("\n=== uv配置优化建议 ===")
    print("1. 设置国内镜像源以提高下载速度:")
    print("   uv config set pip.index-url https://pypi.tuna.tsinghua.edu.cn/simple")
    print("   uv config set pip.extra-index-url https://download.pytorch.org/whl/cu131")
    
    print("\n2. 启用并行下载:")
    print("   uv config set install.parallel true")
    
    print("\n3. 设置缓存目录:")
    print("   uv config set cache-dir /path/to/your/cache/directory")

def main():
    """主函数"""
    benchmark_uv_vs_conda()
    optimize_uv_config()

if __name__ == "__main__":
    main()

# 清理conda环境（可选）
conda clean --all --yes

# 卸载conda（可选）
# Windows: 控制面板 -> 程序和功能 -> 卸载Anaconda
# Linux/macOS: rm -rf ~/anaconda3

# 安装uv
pip install uv

# 设置国内镜像
test -f ~/.config/uv/uv.toml || mkdir -p ~/.config/uv
cat > ~/.config/uv/uv.toml << EOF
[package_index]
index_url = "https://pypi.tuna.tsinghua.edu.cn/simple"
extra_index_url = [
    "https://download.pytorch.org/whl/cu131"
]
EOF

# 创建并使用uv虚拟环境
uv venv
uv activate

#!/bin/bash

# 检查当前环境
echo "=== 检查当前环境 ==="
python --version

# 安装uv
echo "\n=== 安装uv ==="
pip install uv

# 创建并激活虚拟环境
echo "\n=== 创建虚拟环境 ==="
uv venv
uv activate

# 设置国内镜像
echo "\n=== 设置国内镜像 ==="
if [ -d "~/.config/uv" ]; then
    mkdir -p ~/.config/uv
fi
cat > ~/.config/uv/uv.toml << EOF
[package_index]
index_url = "https://pypi.tuna.tsinghua.edu.cn/simple"
extra_index_url = [
    "https://download.pytorch.org/whl/cu131"
]
EOF

# 安装PyTorch GPU版
echo "\n=== 安装PyTorch GPU版 ==="
uv pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu131

# 安装常用依赖
echo "\n=== 安装常用依赖 ==="
uv pip install numpy pandas matplotlib scikit-learn jupyter

# 验证安装
echo "\n=== 验证安装 ==="
python -c "
import torch
import numpy
import pandas
import matplotlib
import sklearn
print('PyTorch版本:', torch.__version__)
print('CUDA可用:', torch.cuda.is_available())
if torch.cuda.is_available():
    print('GPU名称:', torch.cuda.get_device_name(0))
print('numpy版本:', numpy.__version__)
print('pandas版本:', pandas.__version__)
print('matplotlib版本:', matplotlib.__version__)
print('scikit-learn版本:', sklearn.__version__)
"

# 测试性能
echo "\n=== 测试性能 ==="
python -c "
import torch
import time

# 创建测试数据
a = torch.randn(1024, 1024, device='cuda')
b = torch.randn(1024, 1024, device='cuda')

# 测试矩阵乘法
start = time.time()
for _ in range(100):
    c = torch.matmul(a, b)
torch.cuda.synchronize()
end = time.time()
print(f'GPU矩阵乘法100次耗时: {end - start:.4f}秒')
"

echo "\n=== 安装完成 ==="