Skip to content
Other

Laya DOM Full XL

файнтюн для сопоставления DOM-элементов

2 min read

Исследовательский full-encoder файнтюн движка Laya для попарного сравнения DOM-элементов. На вход подаются baseline и candidate версии веб-страницы, на выходе — решение EQUIVALENT или NOT_EQUIVALENT. База — мультиязычный convaiinnovations/laya с энкодером mmBERT-base на архитектуре ModernBERT, 321.9M параметров.

Ключевые особенности

  • Сравнивает два описания DOM-элемента и определяет, один ли это функциональный элемент для пользователя.
  • Энкодер ModernBERT: 22 слоя, hidden size 768, словарь 256k; голова решений — 2 transformer-слоя со scorer.
  • 321.9M параметров, в этом чекпоинте обучаются все; model.safetensors — state dict Laya RL-agent, а не обычный save_pretrained из Transformers.
  • Обучение на 9 000 синтетических пар: 7 217 train / 1 783 held out, RLCD, 4 эпохи, bf16, max sequence length 1024.
  • Это не универсальный DOM-парсер и не полная система восстановления локаторов — только исследовательский чекпоинт.

Пример

import laya

agent = laya.load("morozovmisha/laya-dom-full-xl-s42", device="cpu")
answer = agent.predict(state, questions, lang="en")
choice = answer["answers"]["is_match"]["choice"]
verdict = "EQUIVALENT" if choice == "same_element" else "NOT_EQUIVALENT"
ПодробнееAccuracy: 98.37% на held-out synthetic Generator A XL (1 783 пары), 96.75% на реальных страницах (770 пар, 13 сайтов), 90.78% на Generator B.
Полный чекпоинт рекомендован для transfer; parameter-efficient top-4 вариант на реальных страницах даёт 42.21%.
Обучение: 1× RTX A4000 16 GB, около 26 минут.
Ограничения: небольшая и коррелированная выборка реальных страниц, нет замеров калибровки и абляций, не тестировались shadow DOM и динамические страницы. Лицензия Apache-2.0, энкодер mmBERT-base — MIT.
Discuss Telegram 0 17

Will it run?

Runs in F16

Fits without quantisation.

  • F16weights 0.6 GB · ready files1.2 GB
  • FP8 / INT8recommendedweights 0.3 GB0.8 GB
  • 4-bitweights 0.2 GB0.7 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 · F16 — 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.

    # morozovmisha/laya-dom-full-xl-s42
    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 morozovmisha/laya-dom-full-xl-s42 --local-dir ~/models/laya-dom-full-xl-s42
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to laya-dom-full-xl-s42
  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 laya (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

morozovmisha/laya-dom-full-xl-s42
Author
morozovmisha
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

170

Ask AI about this model

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

Comments

Other

PRISM

PRISM
Other

LaUER

LaUER