Tencent open-sources EVIE visual document retrieval models, hitting 66.75 nDCG@10 on ViDoRe V3
jacek2023 · reddit · 2026-09-07
Tencent released EVIE-8B and EVIE-4.5B high-capacity visual document retrieval models on Hugging Face, scoring 66.75 and 66.02 nDCG@10 on ViDoRe V3. Key points:
- 8B model outputs 4096D per-token multi-vector embeddings preserving layout, typography, charts and table structures
- 4.5B uses Prefix-MRL: a single 2048D projection truncatable at runtime to 64–2048 dimensions without separate models
- Training-free HAC clustering compresses 750 tokens/page to 32 vectors, cutting index storage to 3.81 GiB per million pages
- EVIE-ARD distillation recipe reproduces student training from the 8B teacher; validated across 138 multilingual tasks on ViDoRe V1/V2/V3 and JinaVDR
More from Models
- Local benchmark: Flint matches Qwen3.6 quality at 50/52 while running 3.6x faster on M4 Pro — RichardSocher · 2026-10-27
- ICML 2026 position paper: LLM 'thinking traces' are a harmful anthropomorphism — rao2z · 2026-10-27
- One stray 'u' and Claude starts spelling like a Brit — cantrell · 2026-10-27
- Cohere's Invent-A-Dataset paper: zero-seed synthetic data beats frontier APIs on quality and diversity — idanbeck · 2026-10-27
- ARC Prize redesigns leaderboard: highest reasoning effort by default, CSV export — mhmazur · 2026-10-27
- Cohere: open-weight Qwen3.8-Flash-Next tried uploading probe file to Beijing cloud bucket in sandbox — wstone_bd · 2026-10-27