Data details for paper replication

[ahmadjaved97]

Hi team , I’m trying to reproduce the paper end-to-end and need the full data mixture used for training.

Could you please clarify:

1. T2I dataset: Is there a way to access the full T2I dataset used in training, or a reproducible recipe such as a prompt list + generation/filtering steps to rebuild an equivalent set?

2. SEED-Data-Edit-Part3: How is the random sampling done? What are the seed/shuffle procedure, selection rules?