DistilVDR: A 524M Compact Visual Document Retriever Distilled from 8B
nanovdr · hf · 2026-08-12
DistilVDR is a compact 524M parameter visual document retriever distilled from an 8B teacher model. Using a dual-student distillation approach with cosine alignment, it achieves near-teacher accuracy without requiring relevance labels, significantly reducing index size and improving retrieval speed.
More from Research
- Why AI Needs Memory Engineering: Graph Memory Beats Flat Vector Stores — anirbanbandyo · 2026-08-12
- LlamaIndex Launches ExtractBench: 4,869 Pages of Complex Docs Across 8 Domains — llama_index · 2026-08-12
- Mitsuhiko Calls for More Direct Criticism Over Academic 'Glass Hearts' — cloneofsimo · 2026-08-12
- Beyond 'Slop': The Rise of Interactive AI Music Experiences — jordiponsdotme · 2026-08-12
- Decoding-Level Taboo: A Diagnostic Stress Test for LLM Robustness — Tadanobu Chuyo Kamijo · 2026-08-12
- AI Memory Benchmarks Under Fire: Scoring Metric Alone Causes 24-Point Gap — True_Mongoose_7073 · 2026-08-12