Skip to content
Other

VultronRetrieverPrime-Qwen3.5-8B-v2

Flagship visual document retriever

2 min readupdated

The model solves the task of visual search across documents, scans, tables, charts, and reports in six languages simultaneously (English, French, German, Spanish, Italian, and Portuguese). Built on the Qwen/Qwen3.5-9B hybrid architecture (8.4 billion parameters), it uses a late-interaction approach (ColBERT-style), scoring pages by their rendered images using 4096-dimensional multi-vector embeddings. Unlike simple text embeddings, this approach enables precise matching of queries with tables, charts, and layout elements. The model is useful for multimodal RAG system developers who need maximum search accuracy over complex documents.

Key features

  • ViDoRe V3: 67.46 average nDCG@10 across 8 public tasks, outperforming many alternatives.
  • Bidirectional attention: all 8 full-attention layers operate bidirectionally for better context understanding.
  • 4096-dim multi-vectors: detailed token-level representation (up to 1792 visual tokens per page).
  • Efficiency: requires about 17 GB of VRAM in bf16 and runs on a single GPU.

What you can do

  • Multimodal RAG: integrate into a search system or chatbot for accurate answers retrieval from PDF scans, reports, and slides.
  • Table and diagram analysis: extract information from pages with complex layouts, charts, and tabular data.
  • Multilingual search: index documents in six European languages without quality loss.
  • Integration via vLLM: deploy in production using a pooling runner and the MaxSim pipeline.

Example

Model initialization and encoding queries with images using the Sentence Transformers library:

from sentence_transformers import MultiVectorEncoder

model = MultiVectorEncoder("vultr/VultronRetrieverPrime-Qwen3.5-8B-v2")

queries = [
    "What is the variable represented on the y-axis of the graph?",
    "Total outlay is maximum in which year?",
]
images = [
    "[https://huggingface.co/datasets/sentence-transformers/example-documents/resolve/main/doc1.jpg](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/doc2.jpg)",
]

query_embeddings = model.encode_query(queries)
image_embeddings = model.encode_document(images)

scores = model.similarity(query_embeddings, image_embeddings)
print(scores)
Read moreDetails
• Benchmarks: ViDoRe V3 (67.46 nDCG@10), MTEB V1 (93.35), V2 (72.91).
• Requirements: about 17 GB VRAM (bf16), transformers library support >= 5.2.0.
• Limitations: retriever only, requires a separate reader/LLM for response generation; supports exactly 6 languages.
Discuss Telegram 1 28

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 16 GB · ready files18 GB
  • FP8 / INT8recommendedweights 7.8 GB9.1 GB
  • 4-bitweights 4.4 GB5.3 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 3080 · BF16 — from RTX 4090

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # vultr/VultronRetrieverPrime-Qwen3.5-8B-v2
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download vultr/VultronRetrieverPrime-Qwen3.5-8B-v2 --local-dir ~/models/VultronRetrieverPrime-Qwen3.5-8B-v2
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

  1. Qwen3.5-9B-Basefamily root
  2. Qwen3.5-9B
  3. VultronRetrieverPrime-Qwen3.5-8B-v2

New posts about versions of Qwen3.5-9B-Base (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 6All on Hugging Face
Merges 1All on Hugging Face

About the model

vultr/VultronRetrieverPrime-Qwen3.5-8B-v2
Author
vultr
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

281

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments