Skip to content
Computer vision

WeVisOne-VL-4B

объекты и пиксели в одной VLM

1 min read

WeVisOne-VL-4B от WeChat Vision (Tencent) — унифицированная vision-language модель на базе Qwen3-VL. Она объединяет object-level и pixel-level восприятие через structured next-token prediction и лёгкий pixel decoder на 9,6 млн параметров; всего в модели 4,83 млрд параметров. Предназначена для исследований visual grounding, сегментации, OCR-локализации и image matting.

Ключевые особенности

  • Open-vocabulary object detection, referring localization, object counting, object pointing и OCR.
  • Referring segmentation и promptable alpha matting.
  • Специальные coordinate tokens <0>…<1000> для нормализованных боксов и точек; если объекта нет, возвращается None.
  • После каждого бокса для dense-задач выдаётся instance token <OBJ> (token ID 152678).
  • Обязательный mask_decoder.pt на 9 609 764 параметра: visual adapter, box-guided query decoder и dense prediction head.

Пример

from PIL import Image
from wevisone_vl import WeVisOneVL

model = WeVisOneVL("tencent/WeVisOne-VL-4B", device="auto", dtype="bfloat16")
image = Image.open("example.jpg").convert("RGB")
result = model.predict(image, task="matting", query="person", temperature=0.0)
print(result.response, len(result.masks))
ПодробнееВ модели 4,83 млрд параметров; также выпущен вариант WeVisOne-VL-2B.
Инференс идёт без явного CoT и рассчитан на single-turn восприятие изображения; video grids pixel decoder отклоняет.
Границы alpha чувствительны к разрешению, сильной окклюзии, сложной прозрачности и формулировке промпта.
Код Tencent в репозитории — под Apache License 2.0.
Discuss Telegram 94 9.8K

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 9.0 GB · ready files10 GB
  • FP8 / INT8recommendedweights 4.5 GB5.4 GB
  • 4-bitweights 2.5 GB3.3 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 2060 · BF16 — from RTX 2080 Ti

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.

    # tencent/WeVisOne-VL-4B
    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 tencent/WeVisOne-VL-4B --local-dir ~/models/WeVisOne-VL-4B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to WeVisOne-VL-4B
  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

New posts about versions of WeVisOne-VL-4B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 4All on Hugging Face
Fine-tunes 1All on Hugging Face

About the model

tencent/WeVisOne-VL-4B
Author
tencent
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

9.8K94

Ask AI about this model

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

Comments

Computer vision

OvisOCR3

OvisOCR3