Skip to content
Other

UltraBERT Curation

380M-параметровый специалист по курации данных

2 min read

davanstrien выпустил компактную модель решений для очистки и сортировки данных: она принимает текст и набор именованных вопросов, а возвращает вероятности по вариантам без генерации текста. В основе — энкодер UltraBERT-380M и scoring head GLiClass; модель говорит на Jev / System One API и работает как обычный GLiClass zero-shot классификатор.

Ключевые особенности

  • Обучена на 535 730 строках; открыто 516 964 — у каждой строки указаны источник, разметчик и лицензия.
  • Поддерживает типы вопросов choice, noul и score, а также GLiClass pipeline.
  • Для курации веб-страниц, промптов, математических задач и библиотечных записей: тема, формат, образовательная ценность, сложность, тип задачи, валидность, каталожный класс.
  • До ~530 question-rows/s на одной RTX PRO 6000 через packed GPU batch scoring.
  • Скрипт curate.py запускает оценку датасета на Hugging Face Jobs одной командой.
  • Это специалист, а не универсальная модель: на общих решениях она заметно слабее.

Пример

import torch
from gliclass import GLiClassModel, ZeroShotClassificationPipeline
from transformers import AutoTokenizer
model = GLiClassModel.from_pretrained("davanstrien/ultrabert-380m-curation", dtype=torch.bfloat16)
tokenizer = AutoTokenizer.from_pretrained("davanstrien/ultrabert-380m-curation")
pipe = ZeroShotClassificationPipeline(model, tokenizer, classification_type="single-label", max_length=2048, device="cuda:0")
pipe("…текст страницы…", ["News", "Tutorial", "Product page", "Forum thread"], prompt="What is the format of this web page?")
ПодробнееCuration score: 0.705 против 0.500 у GLiClass Edge и 0.582 у d1-2B.
На общих решениях Decision Index 22.1 против 47.5 и 51.1 — это действительно специалист.
API принимает до 4 096 токенов, быстрый путь обрезает на 2 048; тестировалась модель только на английском.
Выходы стоит воспринимать как сигналы для сортировки, а не как истину.
Discuss Telegram 12 283

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 1.4 GB · ready files2.1 GB
  • BF16weights 0.7 GB1.3 GB
  • FP8 / INT8recommendedweights 0.4 GB0.9 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 · F32 — 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.

    # davanstrien/ultrabert-380m-curation
    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 davanstrien/ultrabert-380m-curation --local-dir ~/models/ultrabert-380m-curation
    # 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 UltraBERT-380M (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 3All on Hugging Face
Fine-tunes 12All on Hugging Face

About the model

davanstrien/ultrabert-380m-curation
Author
davanstrien
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

28312

Ask AI about this model

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

Comments