How to replicate inlet condition in rollout in Learning to Simulate paper .

**Description:**

We are working on a project to simulate fluid flow using a neural network. We're facing challenges in accurately modeling the inflow of fluid into the domain. We are replicating the paper Learning to Simulate .

**Specific Questions:**

1. How can we effectively implement inlet boundary conditions at time of rollout ?
2. If we add particles position like we are doing in simple cases dimension of tensor will not match , How we can ensure about that

Any insights, code examples, or references to relevant literature would be greatly appreciated.