Skip to content
Computer vision

d1-2B

компактная модель решений для текста, картинок и видео

1 min read

LiquidAI выпустила d1-2B — компактную модель решений на базе LFM3-VL-2B. Это не чат-модель: она отвечает на именованные вопросы о состоянии (текст или JSON), изображениях и видео, возвращая типизированные ответы с вероятностями без генерации токенов.

Ключевые особенности

  • 1,92 млрд параметров, контекст 128K токенов; поддерживает 16 языков, включая русский.
  • Один forward pass: можно задать любое число вопросов об одном состоянии за вызов.
  • Типы вопросов: yes/no, choice, score, извлечение text/number/date со span, оценка числа с интервалом.
  • Ищет объекты на изображениях рамками и считает их; принимает видео как набор кадров.
  • По данным карточки, лидирует среди открытых d1 в среднем по более чем 100 бенчмаркам решений, обгоняя d1-3B при размере менее двух третей.

Пример

import torch
from transformers import AutoModel
model = AutoModel.from_pretrained("LiquidAI/d1-2B", trust_remote_code=True, dtype=torch.bfloat16).to("cuda")
q = {"refund": {"type": "noul", "instructions": "Refund?"}, "team": {"type": "choice", "instructions": "Team?", "criteria": {"billing": "Charges", "shipping": "Delivery"}}}
out = model.system_one("Charged twice; parcel missing.", q)
print(out["answers"])
ПодробнееСкорость: 5,7 мс на решение на NVIDIA H100 и 3,8 мс на AMD MI325X.
Надёжность ответов падает после примерно 64K токенов состояния.
Оценки надёжны для подсчётов и баллов, хуже для свободных величин.
Лицензия LFM Open License v1.0; знания — не сильная сторона, лучше подходит для routing, модерации, извлечения и визуальной проверки.
Discuss Telegram 201 6K

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 3.6 GB · ready files4.4 GB
  • FP8 / INT8recommendedweights 1.8 GB2.5 GB
  • 4-bitweights 1.0 GB1.6 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.

    # LiquidAI/d1-2B
    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 LiquidAI/d1-2B --local-dir ~/models/d1-2B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to d1-2B
  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

  1. LFM3-1.7B-Basefamily root
  2. LFM3-VL-2B
  3. d1-2B

New posts about versions of LFM3-1.7B-Base (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 9All on Hugging Face

About the model

LiquidAI/d1-2B
Author
LiquidAI
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

6K201

Ask AI about this model

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

Comments