Perplexity Invites Retrieval Model Submissions for Q2D-Web Evaluation
Key Info
Perplexity is inviting researchers and developers to submit publicly available Hugging Face retrieval models for evaluation, with the full technical report now available.
Highlights
- The evaluation uses the Q2D-Web benchmark, which combines three relevance sets: agent citations, production web rankings, and LLM-expanded judgments.
- Earlier results assessed 13 retrieval models using Recall@1000, with pplx-embed-v1-4b leading in both web and combined rankings.