Skip to content
Other

MAI-UI-2B5

GUI-агент для реальных интерфейсов

2 min readupdated

MAI-UI — семейство фундаментальных GUI-агентов размерами 2B, 8B, 32B и 235B-A22B; в репозитории представлен вариант 2B5 на 2,13 млрд параметров. Модель предназначена для навигации и выполнения задач в графических интерфейсах, включая мобильные сценарии и взаимодействие с пользователем. В карточке описаны self-evolving data pipeline, native device–cloud collaboration и online RL.

Ключевые особенности

  • SOTA на GUI grounding: 73.5% ScreenSpot-Pro, 91.3% MMBench GUI L2, 70.9% OSWorld-G и 49.2% UI-Vision.
  • Мобильная навигация: 76.7% на AndroidWorld и 41.7% success rate на MobileWorld.
  • Поддержка native agent–user interaction и MCP tool calls через расширенный data pipeline.
  • Device–cloud collaboration выбирает on-device или cloud execution по состоянию задачи и чувствительности данных.
  • Online RL: масштабирование параллельных сред с 32 до 512 даёт +5.2 пункта, увеличение step budget с 15 до 50 — +4.3.

Пример

Сценарий из карточки: агенту нужно выполнить задачу в мобильном приложении. Он анализирует состояние GUI, при необходимости вызывает MCP-инструмент, а система device–cloud решает, выполнять шаг локально или в облаке — в зависимости от состояния задачи и чувствительности данных.

ПодробнееБенчмарки: ScreenSpot-Pro — 73.5%, AndroidWorld — 76.7%, MobileWorld — 41.7%.
Device–cloud collaboration повышает on-device производительность на 33% и сокращает облачные API-вызовы более чем на 40%.
Семейство включает варианты 2B, 8B, 32B и 235B-A22B; требования к VRAM в карточке не указаны.
Примеры
Discuss Telegram 0 18

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.

    # Yab762/MAI-UI-2B5
    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 Yab762/MAI-UI-2B5 --local-dir ~/models/MAI-UI-2B5
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to MAI-UI-2B5
  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 MAI-UI-2B5 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 1All on Hugging Face

About the model

Yab762/MAI-UI-2B5
Author
Yab762
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

180

Ask AI about this model

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

Comments