Inquiry about the Implementation of “Self-Correction” in Fig. 7

[ashun989]

Thank you very much for your excellent work!
I was wondering if you could kindly share the implementation details of the Self-Correction shown in Fig. 7.
Additionally, does the training of dLLM-Var enhance this capability compared to the baseline model (e.g. LLaDA)?

[maomaocun]

Thank you for your question. We've trained a model with editing capabilities, but it hasn't achieved the best standards across all tasks, so we've only included corresponding cases in the paper without full results. Currently, this model is scheduled for further training and should be released in about half a month.

[wjczf123]

I also find this section extremely unclear. I have no idea how the author implemented this function.