Skip to content
Computer vision

Qwen3.5-4B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING

4B-модель без отказов

1 min readupdated

Это файнтюн Qwen3.5-4B на 4,54 млрд параметров от ivannapoleon54321, сделанный через Unsloth на четырёх датасетах Claude-4.6-OS. Модель мультимодальная (image-text-to-text): по карточке, поддерживает thinking, работу с изображениями и полностью uncensored-режим. Языки — английский и китайский.

Ключевые особенности

  • Без отказов: 4 отказа из 100 против 94 у оригинальной Qwen3.5-4B; KL divergence — 0.0406.
  • Vision: автор отдельно отмечает, что работа с изображениями протестирована после дообучения.
  • Исправленный Jinja-шаблон: борется с повторами, затяжным thinking и циклами, улучшает обработку tool calls.
  • Кванты: рекомендуются минимум q4ks (non-imatrix) или IQ3S (imatrix); контекст по умолчанию — 256k.
  • Бенчмарки: на части тестов авторского сравнения модель выше Qwen3.5-4B-Instruct, но не на всех.

Пример

Подайте изображение и запрос вроде: «Опиши, что на картинке, и предложи три короткие подписи для соцсетей. Думай пошагово, но без лишних рассуждений». Модель должна вернуть описание и варианты подписей в thinking-режиме — vision в карточке заявлен как рабочий.

ПодробнееБенчмарки: ARC 0.566, ARC/e 0.766, BoolQ 0.873. У Qwen3.5-4B-Instruct — 0.505, 0.688 и 0.892 соответственно, то есть не все метрики выше базы.
Обучение: Unsloth, 4 датасета Claude-4.6-OS, локальное железо; модель обучена после «Heretic»-децензурирования.
Ограничения: видео-часть не тестировалась; языки — en/zh; контекст 256k по умолчанию.
Discuss Telegram 0 18

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 8.5 GB · ready files9.8 GB
  • FP8 / INT8recommendedweights 4.2 GB5.2 GB
  • 4-bitweights 2.4 GB3.1 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.

    # ivannapoleon54321/Qwen3.5-4B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING
    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 ivannapoleon54321/Qwen3.5-4B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING --local-dir ~/models/Qwen3.5-4B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING
    # 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

  1. Qwen3.5-4B-Basefamily root
  2. Qwen3.5-4B
  3. Qwen3.5-4B-heretic
  4. Qwen3.5-4B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING

New posts about versions of Qwen3.5-4B-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 12All on Hugging Face
Merges 1All on Hugging Face

About the model

ivannapoleon54321/Qwen3.5-4B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING
Author
ivannapoleon54321
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

180

Ask AI about this model

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

Comments