Skip to content
Computer vision

Qwen2.5-VL-3B Heretic

decensored VLM для картинок и видео

1 min readupdated

Это расцензуренная версия Qwen2.5-VL-3B-Instruct, полученная через Heretic v2.1.0. Модель на 3,75 млрд параметров понимает изображения и видео, отвечает на вопросы, извлекает структурированные данные и может работать визуальным агентом. В карточке указаны параметры ARA и возможность воспроизведения результата.

Ключевые особенности

  • Снижены отказы: 3/100 против 99/100 у оригинальной Qwen2.5-VL-3B-Instruct по оценке из карточки.
  • Arbitrary-Rank Ablation затрагивает слои 14–31 и сохраняет полезное поведение с заданными весами.
  • Разбирает текст, графики, иконки, таблицы и документы; умеет выдавать bounding boxes, points и JSON с координатами.
  • Понимает видео длиной более часа, использует динамический FPS и умеет находить нужные фрагменты.
  • Может выступать визуальным агентом для computer use и phone use; ScreenSpot — 55.5.
  • Результат воспроизводим: инструкция лежит в каталоге reproduce.

Пример

from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    "Kasemcyber/Qwen2.5-VL-3B-Instruct-UNCENSORED-Heretic",
    torch_dtype="auto", device_map="auto"
)
processor = AutoProcessor.from_pretrained("Kasemcyber/Qwen2.5-VL-3B-Instruct-UNCENSORED-Heretic")
ПодробнееОтказы: 3/100 против 99/100 у оригинала.
KL divergence: 0.3007.
Нужен свежий transformers: pip install git+https://github.com/huggingface/transformers accelerate, иначе возможен KeyError: 'qwen2_5_vl'.
Отдельных требований к VRAM в карточке нет; параметров — 3,75 млрд.
Discuss Telegram 0 62

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 7.0 GB · ready files8.2 GB
  • FP8 / INT8recommendedweights 3.5 GB4.3 GB
  • 4-bitweights 2.0 GB2.7 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 3080

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.

    # Kasemcyber/Qwen2.5-VL-3B-Instruct-UNCENSORED-Heretic
    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 Kasemcyber/Qwen2.5-VL-3B-Instruct-UNCENSORED-Heretic --local-dir ~/models/Qwen2.5-VL-3B-Instruct-UNCENSORED-Heretic
    # 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 Qwen2.5-VL-3B-Instruct (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 7All on Hugging Face

About the model

Kasemcyber/Qwen2.5-VL-3B-Instruct-UNCENSORED-Heretic
Author
Kasemcyber
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

620

Ask AI about this model

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

Comments