Different reward on local and remote environments

[orlov_alexander]

We have same issues. Our RL model could score 100+ on local, and score like 14 on remote.