Request for Open-Sourcing the GPT-Based Evaluation Pipeline

[Zhangshy1019]

Hi, thanks for the great work!

I noticed that the paper/project uses GPT-based evaluation for assessing generation quality (e.g., realism, consistency, interaction quality, etc.).

Would it be possible to release the evaluation pipeline/scripts used in the paper?

In particular, it would be very helpful to have:

• GPT prompts/templates
• evaluation scripts
• score parsing logic
• API settings (model version, temperature, etc.)
• batch evaluation pipeline

This would greatly improve reproducibility and help the community perform fair comparisons with future methods.

Thanks!