Because only the following versions of torch{sys_platform == 'linux'} are available:

[baiyajin]

(.venv) a@DESKTOP-HAVM426:~/WeClone$ uv pip install --group main -e .
Updated https://github.com/hiyouga/LLaMA-Factory.git (50780499edce9217c857ebcd6402371aa3fe3c44)
× No solution found when resolving dependencies:
╰─▶ Because only the following versions of torch{sys_platform == 'linux'} are available:
torch{sys_platform == 'linux'}<2.7.1
torch{sys_platform == 'linux'}>=2.7.1+cu126
and torch==2.7.1+cu126 depends on system:cuda==12.6, we can conclude that torch{sys_platform == 'linux'}==2.7.1 depends on system:cuda==12.6.
And because triton==3.3.1 depends on system:cuda==12.9, we can conclude that torch{sys_platform == 'linux'}==2.7.1 and triton{sys_platform == 'linux'}==3.3.1 are incompatible.
And because you require torch{sys_platform == 'linux'}==2.7.1 and triton{sys_platform == 'linux'}==3.3.1, we can conclude that your requirements are unsatisfiable.

[MapleWithered]

#198

Fixing triton's index to pytorch-cu126 solves this problem.

Adding these lines under [tool.uv.sources] section in pyproject.toml

triton = [
  { index = "pytorch-cu126", marker = "platform_system == 'Windows'" },
  { index = "pytorch-cu126", marker = "platform_system == 'Linux'" },
]