Inconsistent IRFs: Mixing log() and Level Equations in Dynare Model Block?

The mixing of logs and levels seems fine. What you report may not be too unusual. It has been discussed e.g. at Positive monetary policy shock with nominal interest rate falling?