Skip to content
Computer vision

Qwen3.5-9B CyberSecurity (Abliterated)

a cybersecurity model without unnecessary refusals

2 min readupdated

This multimodal model (takes images and text as input, outputs text) is built for vulnerability analysis and information security tasks. It uses unsloth/Qwen3.5-9B (9.41 billion parameters) as its base, merged with jeonjw's Qwen3.5-9B-CyberSecurity-LoRA adapter (a lightweight fine-tuning add-on). Its standout feature is abliteration (removing the refusal direction vector from the model's weights to prevent standard refusals on security-related queries). It is a solid fit for security researchers, pentesters, and developers working in authorized environments.

Key features

  • Adapter merge: The specialized cybersecurity LoRA is merged into the base model in fp16 format.
  • Weight abliteration: The refusal direction was computed via activation differences and projected out of the self_attn.o_proj, linear_attn.out_proj (in 24 of the 32 hybrid attention layers), and mlp.down_proj matrices at scale = 1.0.
  • Intact vision: The vision tower remains untouched, keeping the model's multimodal capabilities intact.
  • Checkpoint details: The MTP (Multi-Token Prediction) head is missing, so converting to GGUF requires the --no-mtp flag (a pre-built GGUF release is also available separately).

What you can do

  • Break down how specific vulnerabilities work (such as format string bugs) and get remediation guidance.
  • Power local assistants for code security reviews and authorized penetration testing.
  • Run it on local hardware via transformers or grab the standalone GGUF build for local runners.

Example

import torch
from transformers import AutoModelForImageTextToText, AutoProcessor

model_id = "jeonjw/Qwen3.5-9B-CyberSecurity-Abliterated"
processor = AutoProcessor.from_pretrained(model_id)
model = AutoModelForImageTextToText.from_pretrained(model_id, dtype=torch.bfloat16, device_map="auto")
messages = [{"role": "user", "content": "Explain how a format string vulnerability works and how to remediate it."}]
inputs = processor(text=[processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True, enable_thinking=False)], return_tensors="pt").to(model.device)
print(processor.decode(model.generate(**inputs, max_new_tokens=512, do_sample=False)[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))
Read moreDetails
Refusals: 17% across 6 offensive security test prompts (using greedy decoding and enable_thinking=False, evaluated by keywords).
Limitations: Abliteration removes only a single direction per layer, so some refusals remain; safety filters are intentionally weakened.
Intended use: Strictly for legitimate research, vulnerability analysis, and defensive work in authorized environments.
Discuss Telegram 1 40

Will it run?

Runs in F16

Fits without quantisation.

  • F16weights 18 GB · ready files20 GB
  • FP8 / INT8recommendedweights 8.8 GB10 GB
  • 4-bitweights 4.9 GB5.9 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 2080 Ti · F16 — from RTX 4090

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.

    # jeonjw/Qwen3.5-9B-CyberSecurity-Abliterated
    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 jeonjw/Qwen3.5-9B-CyberSecurity-Abliterated --local-dir ~/models/Qwen3.5-9B-CyberSecurity-Abliterated
    # 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. Qwen3.5-9B-Basefamily root
  2. Qwen3.5-9B
  3. Qwen3.5-9B
  4. Qwen3.5-9B-CyberSecurity-Abliterated

New posts about versions of Qwen3.5-9B-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 6All on Hugging Face
Merges 1All on Hugging Face

About the model

jeonjw/Qwen3.5-9B-CyberSecurity-Abliterated
Author
jeonjw
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

401

Ask AI about this model

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

Comments