RenderRank reranks documents as images: 35% fewer tokens, beats sub-4B text rerankers
nlpai-lab · hf · 2026-09-29
nlpai-lab introduces RenderRank, a reranker that renders documents as images and scores relevance from compressed visual tokens instead of text sequences. Training first aligns visual relevance scores with a text-based teacher, then refines relative positive/negative scores.
- Across 11 BEIR datasets: 16.5-35.5% fewer input tokens, NDCG@10 of 55.96, outperforming all text baselines under 4B and some larger models
- On four long-document datasets: NDCG@10 of 88.27 with roughly half the input tokens and 1.70x the highest baseline throughput
Results show compressed visual representations are a viable alternative to text tokens for document reranking.
More from Research
- ColNanoVDR distills multi-vector visual document retrieval without documents, keeping 95% NDCG@5 at 149M params — nanovdr · 2026-09-29
- NUS rethinks DiT residual connectivity: 1.73x fewer training iterations, 1.39 FID — NationalUniversityofSingapore · 2026-09-29
- Imprint Reader Decodes Weight Updates into Natural Language, Enables Targeted Edits — Guanxu Chen · 2026-09-29
- SJTU's GeoVerse Synthesizes World-Consistent Novel Views in Geometric Latent Space — SJTU · 2026-09-29
- Tencent Hunyuan Maps Scaling Laws for Encoder-Free Multimodal Pretraining — Tencent-Hunyuan · 2026-09-29
- When Do Model Internals Help? Benchmarking Representation Engineering for LLM Safety — Tianyi Guan · 2026-09-29