Restart vasp calculation. Leave ML_MODE = train unchanged and restart VASP.