Tencent open-sources EVIE-8B and EVIE-4.5B visual document retrieval models
Tencent published EVIE-8B and EVIE-4.5B on Hugging Face, open-sourcing weights, training and distillation pipelines, the HAC token-compression algorithm and evaluation suites under Apache-2.0, with a technical report at arXiv:2610.11553. Both build on a ColQwen3.5 late-interaction multi-vector architecture: EVIE-8B produces 4096-dimensional per-token embeddings, while EVIE-4.5B uses a single 2048D projection whose Prefix-MRL width can be truncated at runtime to 64–2048 dimensions and is distilled from the 8B teacher. The model cards report 66.75 nDCG@10 on ViDoRe V3 for EVIE-8B and 66.02 for EVIE-4.5B, and say HAC compression cuts page representations from roughly 750 vectors to 32, giving about 3.81 GiB of index per million pages; however, a leaderboard table in the same cards lists the two models at 66.24 and 65.70 Mean Task, which does not match those figures.
EVIE's training-free HAC token compression and elastic embedding widths target the index storage cost of visual document retrieval, with the model card citing about 3.81 GiB per million pages at 32 vectors per page.
tencent/EVIE-4.5B Visual Document Retrieval • 5B • Updated 22 days ago • 1.17k • 41
license: apache-2.0 library_name: colpali-engine pipeline_tag: visual-document-retrieval base_model: tencent/EVIE-Preview-4.5B tags:
- colpali-engine
- qwen3_5
- vision-language
- colbert
- late-interaction
- multi-vector
- matryoshka
- vidore
- token-compression
- sentence-transformers datasets:
- vidore/vidore_benchmark
- vidore/vidore_benchmark_v2
- jinaai/jina-vdr
#Tencent#Tencent Youtu Lab#Tencent IMA Product Center#EVIE-8B#EVIE-4.5B#Qwen3.5#ColPali#ViDoRe#JinaVDR#Hugging Face