Skip to content
Computer vision

Llama-3.2-11B-Vision-Instruct WSP/VLGuard

защита от мультимодальных джейлбрейков

1 min readupdated

Это Llama-3.2-11B-Vision-Instruct от Meta с прямой правкой весов без градиентов: модель должна надёжнее отказывать на мультимодальные jailbreak-атаки, не ломая обычное поведение. Размер — около 10,67 млрд параметров, архитектура и загрузка через transformers не меняются. Правку делал LLM-агент Claude Fable 5.1 по статистике активаций, максимум за 500 forward passes.

Ключевые особенности

  • База — meta-llama/Llama-3.2-11B-Vision-Instruct, около 10,67 млрд параметров.
  • Без файнтюна и оптимизатора: 30 rank-1 обновлений в attention-output и MLP-output матрицах слоёв 6–11 и 16–24.
  • Цель — устойчивее отказывать на вредные изображения, опасные инструкции внутри картинки и adversarial-обёртки текста.
  • Остальные тензоры бит-в-бит совпадают с оригиналом; имена, формы, bf16 и шардинг не меняются.
  • В репозитории есть EDIT_REPORT.json со списком правок и их величиной.

Пример

import torch
from transformers import AutoProcessor, MllamaForConditionalGeneration
model = MllamaForConditionalGeneration.from_pretrained(
    'Adeely93/Llama-3.2-11B-Vision-Instruct-WSP-VLGuard-edited',
    dtype=torch.bfloat16, device_map='cuda'
)
processor = AutoProcessor.from_pretrained('Adeely93/Llama-3.2-11B-Vision-Instruct-WSP-VLGuard-edited')
ПодробнееПодробнее
На VLGuard (1 558 записей, greedy decoding): ASR по keyword-судье на всех вредных входах — 0.396 → 0.196; по LLM-судье — 0.471 → 0.283.
Over-refusal на безопасных входах остался 0.000, helpfulness — 0.842 → 0.799.
Ограничения: местами короче описания на безопасных вопросах; выше отказы на безопасных картинках с текстом; описания явных/жестоких изображений всё ещё часто выдаются. Оценка только с greedy decoding.
Discuss Telegram 0 26

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.

    # Adeely93/Llama-3.2-11B-Vision-Instruct-WSP-VLGuard-edited
    # ===== 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 Adeely93/Llama-3.2-11B-Vision-Instruct-WSP-VLGuard-edited --local-dir ~/models/Llama-3.2-11B-Vision-Instruct-WSP-VLGuard-edited
    # 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 Llama-3.2-11B-Vision-Instruct-WSP-VLGuard-edited (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

Adeely93/Llama-3.2-11B-Vision-Instruct-WSP-VLGuard-edited
Author
Adeely93
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

260

Ask AI about this model

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

Comments