Skip to content
Computer vision

deskpaw-CUA-2B

клик по скриншоту за один проход

1 min readupdated

deskpaw-CUA-2B — полностью дообученная 2B vision-language модель от llstar для взаимодействия с GUI. На скриншот и инструкцию она возвращает точку клика за один прямой проход, без цепочки рассуждений и поиска на инференсе. База — Qwen3-VL-2B-Instruct (2.13B), точность bf16.

Ключевые особенности

  • Полный файнтюн Qwen3-VL-2B-Instruct: обновлены все веса, это не адаптер.
  • System-1: один forward pass — сразу координаты целевого клика.
  • Скриншоты до ~5.9M пикселей на входе.
  • Нужен кастомный prediction head: стандартные загрузчики transformers чекпоинт не загрузят.
  • Обучение остановлено рано — на 20 000 из 31 848 шагов.

Пример

# Нужен кастомный код модели; это не drop-in AutoModel.
from custom_modeling import Qwen3VLForConditionalGenerationWithPointer

model = Qwen3VLForConditionalGenerationWithPointer.from_pretrained(
    "deskpaw/deskpaw-CUA-2B", torch_dtype="bfloat16"
)
processor = AutoProcessor.from_pretrained("deskpaw/deskpaw-CUA-2B")
ПодробнееScreenSpot-v2: 87.97% overall (1 272 примера); текст 95.0%, иконки 78.9%.
Лучший результат — step 14 000; выпущенные веса step 20 000 не переоценивались.
Обучение остановлено на 20 000 из 31 848 шагов (~63% одной эпохи).
Нужен кастомный prediction head; кода головы в репозитории нет. Не safety-evaluated; не для автономной работы с реальными системами. Инструкции лучше на английском.
Discuss Telegram 1 100

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 4.0 GB · ready files4.9 GB
  • FP8 / INT8recommendedweights 2.0 GB2.7 GB
  • 4-bitweights 1.1 GB1.7 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.

    # llstar/deskpaw-cua-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 llstar/deskpaw-cua-2b --local-dir ~/models/deskpaw-cua-2b
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to deskpaw-cua-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

llstar/deskpaw-cua-2b
Author
llstar
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

1001

Ask AI about this model

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

Comments

Other

ANT-4B

ANT-4B