Inquiry regarding detailed generation parameters and evaluation metrics for conditional tasks

Hello!
I am currently digging into the conditional generation experiments described in your paper, specifically the tasks involving target chemistry, energy_above_hull, and dft_band_gap. While the main text and Appendix D provide a great overview, I am hoping to get a bit more technical detail to help us better understand the pipeline and apply it to our own research in materials informatics.

Could you please provide some guidance or share data on the following?

1. Detailed Generation Parameters
   Could you share the specific sampling parameters and configurations used for these conditional tasks? For instance, I would love to know the exact configuration format or code snippet used to apply joint conditions, such as {'chemical_system': 'O-Sr-V', 'energy_above_hull': 0.0}.

2. MatterSim Evaluation and Sampling Details
   Could you clarify the exact sampling sizes used across the different chemical systems (e.g., the specific breakdown for ternary and quaternary systems, alongside the 10,240 mentioned for quinary)? Additionally, are the detailed evaluation results—especially the explicit S.U.N.% metrics for these specific tasks—available, perhaps in a tabular format?

Thank you for your time and for sharing this fantastic work with the community!

Start with Appendix D and the conditional-generation experiments for target chemistry, energy_above_hull, and dft_band_gap. Determine whether the sampling configurations, ternary and quaternary sample breakdowns, and S.U.N.% evaluation tables exist in the project or paper materials. Done means documenting or linking the requested parameters and evaluation results.