Possible bug in stress calculation with NEP and DFT-D3

I was trying to calculate elastic constant of silver bulk using nep89 with DFT-D3. The elastic constant never converges with respect to the strain. To investigate further, I checked the consistency of stress and energy. I found that they are not consistent when the system is fully relaxed (ie zero stress), and are only consistent when stress is large (>0.1 eV/A^3).