No official torrent yet β requesting it tells the flock what to host next. You cannot seed a catalog-only page.
Cataloged from tencent/EVIE-Preview-4.5B
All models by tencent βThe most accurate visual document retriever, with native 128-dimensional token vectors.
[blocked]
π Results β’ πΎ Index Cost β’ β‘ Quick Start β’ π¬ Reproducing β’ π§ Architecture β’ π Citation
8 public domains Γ 6 query languages, nDCG@10.
EVIE was trained at 768 visual tokens per page. The 1,792 tier is pure test-time extrapolation β the same weights, never trained or fine-tuned at that budget, and never re-exported. That the model does not merely hold up but gains 0.80 nDCG@10 at more than twice its training budget, improving in 7 of the 8 domains, is a direct read on how well its page representation generalises beyond the resolution it was fit to.
Pick whichever tier fits your compute budget.
The lighter tier holds a million pages in under 180 GiB.
EVIE rows measured with reproduce.sh [blocked]. Comparison rows are the vendors' published ViDoRe V3 public scores.
14 tasks, nDCG@5. First place on the classic boards too.
Tasks 1β10: ViDoRe V1. Tasks 11β14: ViDoRe V2. Board aggregates: V1 91.73 Β· V2 70.87.
Index size is what decides whether multi-vector retrieval actually ships. EVIE emits native 128D token vectors, so the index stays compact at both page budgets.
Scoring stays cheap for the same reason: MaxSim is a late-interaction dot product over the token vectors, so a narrower vector cuts the scoring work exactly as it cuts storage.
Qwen3.5-4B with interleaved GatedDeltaNet linear attention and full attention.Queries in English, French, German, Italian, Spanish, Portuguese and Chinese, retrieving over charts, tables, scientific reports, financial filings and scanned forms β including Japanese-language pages.
ColPali Engine is the reference path β every number on this card comes from it. A Sentence Transformers path is also available for late-interaction pipelines already built on that API.
Or pip install -r requirements.txt if you cloned the repository; that adds pyarrow, which only reproduce.py [blocked] needs.
Self-contained β nothing to clone, no local files to prepare.
β οΈ Apply
enable_bidirectional_attention(model)once after loading, and resetmodel.rope_deltas = Nonebefore every query forward pass. Both are required to reach the scores above β releasedcolpali-engine(through 0.3.17) builds ColQwen3.5 with causal masks, which costs about 1.1 on top-hit MaxSim. The same helper ships asbidirectional.py[blocked] forinfer.pyandreproduce.py.
EVIE also loads as a Sentence Transformers MultiVectorEncoder, exposing the familiar encode_query / encode_document / similarity API with MaxSim scoring built in. Bidirectional attention is baked into the shipped configuration, so no extra call is needed.
MultiVectorEncoder requires Sentence Transformers 6.0.0, which is not on PyPI yet β install from source until it is released:
Both paths above run on the same four example pages, so they are directly comparable. The two sets of scores agree closely; the small differences come from the attention backend and dtype, and the ranking is identical.
Documents may be file paths, URLs or PIL.Image objects. Text passed to encode_document is rendered as a query, since this model has no separate text-document format.
The default page budget is the 768-token tier. To score the 1,792-token tier, raise the pixel budget through processor_kwargs:
Every number on this card is reproducible with the shipped script across all visible GPUs:
On the first run, download_data.py [blocked] fetches the 22 public ViDoRe datasets (~55 GB) from Hugging Face. To reuse an existing directory:
To score the 1,792-token tier directly:
EVIE was trained on approximately 0.8 million high-quality image-query pairs spanning multilingual documents, technical reports, complex financial tables, infographics and document visual QA.
Every mined negative is re-judged by a large multimodal judge before it reaches the loss:
Multi-positive rows are group-aware weighted by 1/positive_count so that positives from the same query never penalise each other in-batch. Rows with empty queries, corrupted images or degraded text are dropped.
π Rank #1 on ViDoRe V3 Β· π Rank #1 on ViDoRe V1+V2 Β· πΎ native 128D token vectors
1,763.58 vectors/page Γ 128 dim Γ 2 bytes Γ 1,000,000 pages Γ· 2^30 = 420.5 GiB Text Query ββββββββΊ ColQwen3_5 (BiDir Attn) ββββββΊ Query Token Embeddings (128D) β Late Interaction (MaxSim) βββΊ Relevance Score βDocument Image ββββββΊ ColQwen3_5 (Dynamic Vision) βββΊ Doc Token Embeddings (128D)pip install "colpali-engine>=0.3.15" accelerateimport torchfrom huggingface_hub import hf_hub_downloadfrom PIL import Imagefrom colpali_engine.models import ColQwen3_5, ColQwen3_5Processor
model_id = "tencent/EVIE-Preview-4.5B"
def enable_bidirectional_attention(model): """Encoder-ize the full-attention layers; the GatedDeltaNet layers stay recurrent.""" for cfg in (model.config, getattr(model.config, "text_config", None)): if cfg is not None: cfg.is_causal = False for module in model.modules(): if module.__class__.__name__ in ("Qwen3_5Attention", "Qwen3Attention"): if hasattr(module, "is_causal"): module.is_causal = False
# 1. Load model and enable bidirectional attentionmodel = ColQwen3_5.from_pretrained( model_id, torch_dtype=torch.bfloat16, device_map="cuda", attn_implementation="flash_attention_2",).eval()enable_bidirectional_attention(model)
# 2. Load processorprocessor = ColQwen3_5Processor.from_pretrained(model_id)
# 3. Prepare inputs β four example document pagespages = [ hf_hub_download("sentence-transformers/example-documents", f"doc{i}.jpg", repo_type="dataset") for i in range(1, 5)]images = [Image.open(p).convert("RGB") for p in pages]queries = [ "What is the variable represented on the y-axis of the graph?", "Total outlay is maximum in which year?",]
image_batch = processor.process_images(images).to(model.device)query_batch = processor.process_queries(queries).to(model.device)
# 4. Generate multi-vector embeddings and scorewith torch.inference_mode(): image_embeddings = model(**image_batch) model.rope_deltas = None # required before query forward query_embeddings = model(**query_batch)
scores = processor.score(query_embeddings, image_embeddings)print(scores)# tensor([[17.3750, 10.9375, 7.8750, 7.3438],# [ 6.5938, 13.3750, 6.2188, 6.0938]])print("Best page per query:", scores.argmax(dim=1))# Best page per query: tensor([0, 1])python infer.py --query "Quarterly revenue report" --image page_1.png --image page_2.pngpip install "sentence-transformers[image] @ git+https://github.com/huggingface/sentence-transformers.git"from sentence_transformers import MultiVectorEncoder
model = MultiVectorEncoder("tencent/EVIE-Preview-4.5B")
queries = [ "What is the variable represented on the y-axis of the graph?", "Total outlay is maximum in which year?",]documents = [ "https://huggingface.co/datasets/sentence-transformers/example-documents/resolve/main/doc1.jpg", "https://huggingface.co/datasets/sentence-transformers/example-documents/resolve/main/doc2.jpg", "https://huggingface.co/datasets/sentence-transformers/example-documents/resolve/main/doc3.jpg", "https://huggingface.co/datasets/sentence-transformers/example-documents/resolve/main/doc4.jpg",]
query_embeddings = model.encode_query(queries)document_embeddings = model.encode_document(documents)print(query_embeddings[0].shape, document_embeddings[0].shape)# torch.Size([23, 128]) torch.Size([755, 128])
scores = model.similarity(query_embeddings, document_embeddings)print(scores)# tensor([[17.3457, 10.8008, 7.8613, 7.3174],# [ 6.5547, 13.3828, 6.2207, 6.0771]])print("Best page per query:", scores.argmax(dim=1))# Best page per query: tensor([0, 1])model = MultiVectorEncoder( "tencent/EVIE-Preview-4.5B", model_kwargs={"attn_implementation": "flash_attention_2", "device_map": "cuda:0"}, processor_kwargs={"size": {"longest_edge": 1792 * 32 * 32, "shortest_edge": 65536}},)bash reproduce.shbash reproduce.sh /path/to/vidoreViDoRe V1 nDCG@5 91.73 (10 tasks)ViDoRe V2 nDCG@5 70.87 (4 tasks)ViDoRe V1+V2 nDCG@5 85.77 (14 tasks)ViDoRe V3 public nDCG@10 64.56 (8 domains x 6 languages, 768 visual tokens)ViDoRe V3 public nDCG@10 65.36 (8 domains x 6 languages, 1792 visual tokens)python -m torch.distributed.run --nproc_per_node=$(nvidia-smi -L | wc -l) reproduce.py \ --boards v3 --max-visual-tokens 1792 --data-root /path/to/vidore@misc{tencent2026evie, title = {EVIE-Preview-4.5B: Rank-1 Multilingual Visual Document Retrieval with 128-Dimensional Multi-Vector Embeddings}, author = {{Tencent}}, year = {2026}, howpublished = {\url{https://huggingface.co/tencent/EVIE-Preview-4.5B}}}