[lambda-labs skill]: reboot the instance first, then check the drivers. After every launch or reboot, verify with nvidia-smi and python -c 'import torch; print(torch.cuda.is_available())' before starting a training run: discovering a dead GPU hours into a job wastes the whole reservation. If the GPU is still missing after a reboot, open a support ticket rather than burning more instance hours.

Context: torch.cuda.is_available() returning False on a Lambda instance is usually a driver state issue.