ml-agents - playing the game with pretrained model

I've created a small game and trained it using ml-agents and PPO algorithm. It trained well, mean reward increased, I can also see the results on tensorboard and the training process was finished. Nevertheless, when I put the .bytes file into 'model' section of the learning brain and run the game, it doesn't play (the model just does nothing) I've also tried to train a model for one of the unity examples on PC, and everything is fine, so there should be my fault somewhere. What could be the problem caused by? I checked the academy, agent, brain, and they look the same. Maybe it's caused by some project settings error? Or observations collecting? There are screenshots in the zip file OS: Windows 10, Unity version:2018.2.16, but I also checked it on 2017.4.18