uv安装Cuda版本的pytorch

修改pyproject.toml

1.[project]的dependencies添加一个torch，官网最新稳定版本2.7.0刚支持cu128(我寻思torchaudio和torchvision应该不用自己手打)

2.添加[tool.uv.sources]和[[tool.uv.index]] ；sources是添加索引pytorch-cu128这个名字反而没那么重要乐意的话随你怎么命名只要记得得和下面name顺手改了就行。index字面意思的索引，cu128是cuda12.8的简写，不同版本改个url就行(友情提示nightly(预览版)版本的url不一样要在whl后面多一层nightly，cuda版本太新就有可能遇到只有预览版支持你的cuda)

3.保存后使用uv sync更新项目依赖就行

    "torch>=2.7.0",
]

[tool.uv.sources]
torch = [
  { index = "pytorch-cu128", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
]
torchvision = [
  { index = "pytorch-cu128", marker = "sys_platform == 'linux' or sys_platform == 'win32'" },
]

[[tool.uv.index]]
name="pytorch-cu128"
url = "https://download.pytorch.org/whl/cu128"
explicit = true