Visual Document Retrieval
Safetensors
sentence-transformers
colpali-engine
qwen3_5
vision-language
colbert
late-interaction
multi-vector
matryoshka
vidore
token-compression
Instructions to use tencent/EVIE-4.5B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- sentence-transformers
How to use tencent/EVIE-4.5B with sentence-transformers:
from sentence_transformers import MultiVectorEncoder model = MultiVectorEncoder("tencent/EVIE-4.5B") queries = ["Which planet is known as the Red Planet?"] documents = [ "Venus is often called Earth's twin because of its similar size and proximity.", "Mars, known for its reddish appearance, is often referred to as the Red Planet.", "Jupiter, the largest planet in our solar system, has a prominent red spot.", ] query_embeddings = model.encode_query(queries) document_embeddings = model.encode_document(documents) similarities = model.similarity(query_embeddings, document_embeddings) print(similarities) - Notebooks
- Google Colab
- Kaggle
CPU-friendly query encoder for EVIE-4.5B
#2
by Ryenhails - opened
Hi, thanks for releasing EVIE, very great work!
We distilled a 150M text-only query encoder for EVIE-4.5B (ColNanoVDR, arXiv:2609.34899). It supports all Prefix-MRL widths and works with existing EVIE indexes, including HAC-compressed ones, so queries can be encoded on CPU. On ViDoRe v3 it keeps 95-96% of EVIE-4.5B's NDCG@10 (93.8-94.9% with HAC).
Model: https://huggingface.co/nanovdr/ColNanoVDR-Q-Ettin150M-EVIE45B-2048-ML
If you find it useful, would you consider linking it to the Huggingface model card? Happy to open a PR. Thanks!
Sure, we'd be very happy! Thanks for your work!