Skip to content
Computer vision

Gemma 4 31B ARA Heretic

decensored-версия базовой Gemma 4

1 min read

richardyoung выпустил decensored-вариант google/gemma-4-31B-it. Это мультимодальная instruct-модель Gemma 4 примерно на 31B параметров; изменения внесены через Heretic v2.0.0 и Arbitrary-Rank Ablation (ARA).

Ключевые особенности

  • Метод: Heretic v2.0.0 + ARA, абляция затрагивает слои 21–56.
  • Снижены отказы: 28/100 против 99/100 у оригинальной модели.
  • KL divergence составляет 0.0845 — это мера отличия от исходной модели.
  • Параметры ARA: preserve_good_behavior_weight 0.2982, steer_bad_behavior_weight 0.0010, overcorrect_relative_weight 0.9045, neighbor_count 13.
  • Воспроизводимость: в репозитории есть каталог reproduce с README.
  • База: Gemma 4 31B Dense, поддержка текста и изображений, контекст до 256K токенов.

Пример

from transformers import AutoProcessor, AutoModelForMultimodalLM

MODEL_ID = "richardyoung/gemma-4-31B-it-ARA-heretic"
processor = AutoProcessor.from_pretrained(MODEL_ID)
model = AutoModelForMultimodalLM.from_pretrained(
    MODEL_ID, dtype="auto", device_map="auto"
)
Подробнее Refusals: 28/100 у этой модели против 99/100 у google/gemma-4-31B-it.
KL divergence: 0.0845; у оригинала — 0 по определению.
ARA: start_layer_index 21, end_layer_index 56; веса 0.2982 / 0.0010 / 0.9045, neighbor_count 13.
Требования к VRAM и железу в карточке не указаны; это 31B-модель, запуск стоит планировать под доступную память.
Это decensored-вариант: снижены отказы, поэтому оценивайте риски при публичном использовании.
Discuss Telegram 0 49

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

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.

    # richardyoung/gemma-4-31B-it-ARA-heretic
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    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 richardyoung/gemma-4-31B-it-ARA-heretic --local-dir ~/models/gemma-4-31B-it-ARA-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 gemma-4-31B-it-ARA-heretic (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

richardyoung/gemma-4-31B-it-ARA-heretic
Author
richardyoung
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

490

Ask AI about this model

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

Comments