Question about a fair comparison between SimBaV2 and standard SAC

[Sen1553]

Hi, thank you for open-sourcing this great project.

I am currently trying to reproduce the results in this repository, and I have a question regarding the comparison between SimBaV2 and standard SAC.

From my reading, the repository seems to implement SAC with the SimBa / SimBaV2 architecture, but I am not sure whether there is also a plain standard SAC baseline implementation in this codebase.

I would like to ask:

1. Does this repository contain a standard SAC implementation without the SimBaV2 architecture?
2. If not, what would be the recommended way to make a fair comparison between SimBaV2 and SAC?
3. For a fair comparison, which settings should be kept the same between the two methods (e.g., training steps, batch size, replay buffer size, update-to-data ratio, entropy tuning, evaluation protocol, number of seeds)?
4. If you have internal baseline results or recommended SAC hyperparameters/codebase for comparison, could you please share them?

My goal is to isolate the effect of the SimBaV2 architecture as clearly as possible.

Thanks in advance for your help.