Skip to content
Other

d3-ultra

мультимодальная модель решений без генерации текста

2 min read

d3-ultra — новая мультимодальная decision-модель от vllm-sr (Decision 3.0, vLLM Semantic Router) с 26,09 млрд параметров, включая 0,46 млрд vision encoder. На вход принимает текст или JSON, опционально изображения и видео, а на выходе даёт вероятности для вариантов ответа: Choice, Yes/No, Score, Box и Track — без генерации текста. Задача в метаданных — zero-shot-classification, лицензия Apache-2.0.

Ключевые особенности

  • Читает вход один раз на запрос и отвечает сразу на все вопросы к нему.
  • Поддерживает несколько изображений и видео в одном запросе; картинки можно вставлять в вопросы и варианты.
  • Box: находит объект по текстовому описанию и возвращает один бокс с confidence.
  • Для видео — бокс в нужный момент или трек по всему клипу с вероятностью видимости цели.
  • Контекст до 131 072 токенов на вопрос; в карточке указан запуск на одной AMD Instinct MI325X.

Пример

from transformers import AutoModel
model = AutoModel.from_pretrained("vllm-sr/d3-ultra", trust_remote_code=True)
out = model.system_one(
    state="Ticket: the invoice total is wrong after the plan change.",
    questions={"route": {"type": "choice", "instructions": "Which team handles this?",
                         "criteria": {"billing": "Billing", "tech": "Technical support"}}},
)
print(out)
ПодробнееМедианное время: 304 мс — 2,16 с на 1–255 вопросов к одному изображению (AMD Instinct MI325X, AutoModel).
Контекст: до 131 072 токенов на вопрос.
Box всегда возвращает ровно один бокс — смотрите на confidence; ниже 0,10 объект считается не найденным.
Лицензия Apache-2.0.
Примеры
Discuss Telegram 51 1.4K

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 49 GB · ready files54 GB
  • FP8 / INT8recommendedweights 24 GB27 GB
  • 4-bitweights 14 GB16 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.

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.

    # vllm-sr/d3-ultra
    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 vllm-sr/d3-ultra --local-dir ~/models/d3-ultra
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to d3-ultra
  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 d3-ultra (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 10All on Hugging Face
Fine-tunes 2All on Hugging Face

About the model

vllm-sr/d3-ultra
Author
vllm-sr
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

1.4K51

Ask AI about this model

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

Comments