[learning_to_simulate] question about generate MPM dataset in GNS as ground trouth.

Dear DeepMind Team,

I'm interested in graph neural network for physics simulation, and now I want to generate some datasets to use as an example to try GNN in structure. When I use the Taichi library to generate the MLS-MPM dataset for fluid, how can I generate the boundary like in the example the solid won't move during the simulation? (for some scenes, the boundary will be influence by the global parameter like gravity, it will move together with fluild)

and another question is, when I generate the dataset from the taichi library, it will output directly a .ply file, I want to transfer the .ply to a .tfrecord file, so which kinds of features of data should be included except for particle position and material id? if the .tfrecord has different attribution from yours, will it also be successfully input?

thanks in advance