Skip to content
Computer vision

Locus-2B

компактный визуальный grounding с проверкой боксов

2 min read

Locus-2B — компактная VLM на базе Qwen/Qwen3-VL-2B-Instruct (1,75 млрд параметров) от MTalha2001. Она находит объект по текстовому описанию и возвращает bounding box в пикселях либо отвечает, что объекта нет на изображении.

Ключевые особенности

  • Referring expression grounding: по изображению и фразе возвращает [x0, y0, x1, y1].
  • Locus-рецепт: изображение префиллится один раз, а запросы идут короткими ветками по кэшу.
  • Box head делает черновик бокса за один проход; затем м��дель параллельно проверяет координатные токены через Jacobi-итерации.
  • В v1.1 умеет отвечать «not in the image»; на gRefCOCO так обработано 60% no-target выражений.
  • v1.1 заметно лучше на мелких объектах: 40.0 Acc@0.5 при 640² против 16.7 у v1.0.
  • Есть CUDA graphs для повторяющихся форм; на RTX A6000 латентность падает до ~98 мс.

Пример

from PIL import Image
from locus.hub import from_pretrained, ground

m = from_pretrained("<org>/Locus-2B", graphs=True)
img = Image.open("street.jpg")
for r in ground(m, img, ["the man in the red shirt", "the bus on the left"]):
    print(r["expression"], r["box"], r["no_target"])
ПодробнееAcc@0.5 — 88.46, Acc@0.9 — 59.43 (verified exact, среднее по 8 сплитам RefCOCO/+/g, 640²).
Мелкие объекты (<2% изображения): 40.0 Acc@0.5 на 640² и 47.8 на 1024².
Нужна CUDA GPU; латентность на RTX A6000: 226 мс без graphs, 98–111 мс с graphs.
Ограничения: RefCOCO-точность чуть ниже v1.0; отвечает не на все no-target запросы; модель только для grounding-запросов.
Discuss Telegram 18 150

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 3.3 GB · ready files4.1 GB
  • FP8 / INT8recommendedweights 1.6 GB2.3 GB
  • 4-bitweights 0.9 GB1.5 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 2060

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.

    # MTalha2001/LOCUS-2B
    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 MTalha2001/LOCUS-2B --local-dir ~/models/LOCUS-2B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to LOCUS-2B
  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 Qwen3-VL-2B-Instruct (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face
Merges 1All on Hugging Face

About the model

MTalha2001/LOCUS-2B
Author
MTalha2001
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

15018

Ask AI about this model

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

Comments