Skip to content
Computer vision

Gemma 4 12B SOMPOA Heresy LoRA

версия без цензуры

1 min read

Это LoRA-адаптер от MuXodious для google/gemma-4-12B-it, сделанный через Heretic v2.2.0.dev0. Базовая модель — 12B Unified из семейства Gemma 4: encoder-free мультимодальная LLM с текстом, изображениями и аудио на входе и контекстом до 256K токенов.

Ключевые особенности

  • Decensored-адаптер: снижает число отказов при небольшом KL divergence относительно базовой модели.
  • Отказы: 11/101 у этой версии против 101/101 у оригинала.
  • KL divergence — 0.0639; в карточке также приведены SOMPOA-параметры, включая direction_index 26.71.
  • Ставится поверх google/gemma-4-12B-it; отдельного кода загрузки LoRA в карточке нет.
  • База понимает текст, изображения и аудио, поддерживает до 256K токенов контекста.

Пример

Сценарий: промпт, на который базовая Gemma 4 12B отвечает отказом, этот LoRA обрабатывает заметно чаще — в замере 11 отказов из 101 против 101 из 101 у оригинала. Для запуска нужна базовая модель google/gemma-4-12B-it и загрузка адаптера.

ПодробнееОтказы: 11/101 против 101/101 у оригинала.
KL divergence: 0.0639.
SOMPOA direction_index: 26.71.
Нужна базовая google/gemma-4-12B-it; адаптер не заменяет её.
Оценка только по отказам и KL — других бенчмарков в карточке нет.
Discuss Telegram 0

Will it run?

Runs in FP8 / INT8

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

  • BF16weights 22 GB · ready files25 GB
  • FP8 / INT8recommendedweights 11 GB13 GB
  • 4-bitweights 6.4 GB7.6 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 5080 · BF16 — from RTX 5090

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

This is a LoRA: counted together with its base model gemma-4-12B-it.

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.

    # MuXodious/gemma-4-12B-it-SOMPOA-heresy-LORA
    # ===== 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 google/gemma-4-12B-it --local-dir ~/models/gemma-4-12B-it
    hf download MuXodious/gemma-4-12B-it-SOMPOA-heresy-LORA --local-dir ~/models/gemma-4-12B-it-SOMPOA-heresy-LORA
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to gemma-4-12B-it
  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

Requires the base model

google/gemma-4-12B-it

This model is a LoRA / adapter. Download the base model first, then load this one on top.

  1. gemma-4-12Bfamily root
  2. gemma-4-12B-it
  3. gemma-4-12B-it-SOMPOA-heresy-LORA

New posts about versions of gemma-4-12B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 4All on Hugging Face

About the model

MuXodious/gemma-4-12B-it-SOMPOA-heresy-LORA
Author
MuXodious
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments