Skip to content
Computer vision

Mage Hand

Mage-VL-Cowork-hands-v2 — GUI-grounding с акцентом на точность клика

1 min read

Файнтюн microsoft/Mage-VL-Cowork от Jordan Newman / angelX для computer use: по скриншоту и описанию цели модель возвращает точку клика. Около 4,74 млрд параметров, текстовая основа — Qwen3-4B-Instruct-2507; LoRA-адаптер уже слит с базой.

Ключевые особенности

  • GUI-grounding: один клик по цели на скриншоте, без type/scroll/drag.
  • Обучение на данных angelX hands: 1 195 шагов, 9 557 примеров (~0,8 эпохи), mean loss 0.0689.
  • Слитый LoRA: r=64, alpha=128, scale 2.0; заменены 252 linear-слоя языковой модели и merger-тензоры.
  • Оценка на 1 258 holdout-строках: target-box hit 94,99%, within ±15 — 94,12%, parsed 100%.
  • Работает с кадрами до 1920×1080; в angelX клик проходит через подтверждение человеком, ошибочный ответ отклоняется.

Пример

Сценарий из карточки: на вход — скриншот до 1920×1080 и описание цели (например, кнопка покупки), на выходе — одна точка клика. В angelX модель только указывает, а клик подтверждает человек; битый или выходящий за кадр ответ отклоняется.

ПодробнееБенчмарк: hands v2 — 94,99% target-box hit и 94,12% within ±15; v1 — 88,08% и 78,70%.
Повтор на RTX 4080: 95,15% box / 93,96% ±15.
Ограничения: holdout ~74% скриптовый/синтетический, native-desktop — всего 12 строк; модель обучена только на клики и не выполняет type/scroll/drag. Offline-точность не равна завершению задачи вживую.
Discuss Telegram 1 44

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

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.

    # Frosty40/mage-hand-Mage-VL-Cowork-hands-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 Frosty40/mage-hand-Mage-VL-Cowork-hands-v2 --local-dir ~/models/mage-hand-Mage-VL-Cowork-hands-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

New posts about versions of mage-hand-Mage-VL-Cowork-hands-v2 (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

Frosty40/mage-hand-Mage-VL-Cowork-hands-v2
Author
Frosty40
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

441

Ask AI about this model

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

Comments

Computer vision

Viper-9B

Viper-9B