Reproduce “Conditional Vendi Score: Prompt-Aware Diversity Evaluation for Generative AI Models and LLMs” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Mohammad Jalali, Azim Ospanov, Amin Gohari, Farzan Farnia

The authors introduce Conditional-Vendi and Conditional-RKE, prompt-aware diversity measures for prompt-conditioned generators. They combine output similarities with corresponding prompt similarities and subtract prompt-only entropy, aiming to distinguish model-induced variation from prompt-driven variation. In Stable Diffusion XL guidance on 10,000 MS-COCO prompts, Conditional-Vendi increased from 29.85 to 32.45 while CLIPScore stayed comparable; scores depend on embedding choice.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.