Skip to content
RCreddit.com·

tencent/EVIE-8B and EVIE-4.5B (High-Capacity Visual Document Retrieval)

AI summary

Tencent has introduced EVIE-8B and EVIE-4.5B, models designed for high-capacity visual document retrieval. These models have been extensively validated across 138 tasks, including ViDoRe V1, V2, V3, and JinaVDR, using four standard metric families: nDCG, Recall, MAP, and MRR u/1. A key feature is the EVIE-ARD Distillation Recipe, which is an anchor-preserving, capacity-aware relation distillation method that reproduces full student training from the 8B teacher model.

Time & source
Published
Sep 7, 2026, 09:47
Source type
Dev community
Tier
Community
Source status
Healthy
Tier is a per-source editorial setting, not a per-item score.

Times shown in UTC

More details
First seenSep 7, 2026, 13:00Time zoneUTC · UTC+0
Article

https://preview.redd.it/3p6234jzk2oh1.png?width=900&format=png&auto=webp&s=b9eab3351d6c6dbd4d5b3fd677a6c40b57f18167

- SOTA Retrieval Performance: 66.75 nDCG@10 on ViDoRe V3, delivering industry-leading visual document retrieval accuracy.

- High-Capacity 4096D Representations: Full per-token multi-vector embeddings preserving fine-grained layout, typography, charts, and table structures.

- Teacher Foundation: Provides capacity-aware relation and margin distillation targets for the lightweight EVIE-4.5B Prefix-MRL model.

- Multi-Benchmark 138-Task Coverage: Thoroughly validated across 138 tasks (ViDoRe V1, V2, V3, and JinaVDR) across 4 standard metric families (nDCG, Recall, MAP, MRR u/1 ).

- Top-Tier Benchmark Performance: 66.75 on ViDoRe V3 for EVIE-8B and 66.02 for EVIE-4.5B with single-projection Prefix-MRL.

- ⚡ Prefix-MRL Elasticity: Single 2048D linear projection. Freely truncate at runtime into ${64, 128, 256, 512, 1024, 2048}$ dimensions without separate models.

- 📦 Ultra-Compact Index (HAC): Training-free Hierarchical Agglomerative Clustering compresses token counts from ~750 down to 32 vectors/page, slashing index storage to 3.81 GiB per million pages.

- 🌐 138 Multilingual Tasks Evaluated: Thoroughly evaluated across ViDoRe V1, V2, V3, and JinaVDR across 4 metric families (nDCG, Recall, MAP, MRR u/1 ).

- 🔬 EVIE-ARD Distillation Recipe: Anchor-preserving, capacity-aware relation distillation reproducing full student training from the 8B teacher.

Source·reddit.com·Full text via RSS