# Environment Checklist

- PyTorch imports successfully
- `torch.cuda.is_available()` is `True`
- At least one CUDA device is visible
- The GPU name matches the machine you expect to be using
- Device capability is printed and recorded
- Triton imports successfully, or you know why it does not
- `torch.version.cuda` is visible when using CUDA-enabled PyTorch
- `nvcc --version` works if you plan to build the CUDA extension
- `nvidia-smi` works if the driver stack is installed

If any line above fails, fix that before working on later tasks.