

RunPod machines are usually being broken

Open a cmd and run below command to see if GPU broken or not



python - <<'PY'
import torch
print("torch", torch.__version__, "cuda", torch.version.cuda)
print("count", torch.cuda.device_count())
print("available", torch.cuda.is_available())
try:
    print("name", torch.cuda.get_device_name(0))
except Exception as e:
    print("name_error", repr(e))
try:
    x = torch.tensor([1.0], device="cuda")
    print("alloc_ok", x)
except Exception as e:
    print("alloc_error", repr(e))
PY