Skip to content
Language models

Vulnerability Detector

finding vulnerabilities in C code and LLVM IR

2 min readupdated

This repository contains two QLoRA adapters (lightweight add-ons on top of a base model using 4-bit quantization) built on Qwen2.5-Coder-7B-Instruct by phazertron as part of a master's thesis at the University of Parma. Both variants are trained on the synthetic SARD Juliet dataset to detect vulnerabilities and classify them by CWE (Common Weakness Enumeration — a standard catalog of software weakness types): the first analyzes C source code, while the second works on LLVM IR intermediate representation. The model outputs a single-line verdict (such as VULNERABLE: CWE-787 or CLEAN). The main finding of the experiment is that the LLVM IR variant is worse at detecting whether a bug is present at all, yet substantially more accurate at naming the specific CWE category. This release is useful for security researchers and ML engineers looking to reproduce the thesis experiments, though it is not meant for auditing real-world codebases.

Key features

  • Two adapters in one repo: the ir and source subfolders hold the variants for LLVM IR (compiled via clang -O0 -emit-llvm) and C source code, respectively.
  • Contrasting metrics: on the test split, the ir variant reached 0.7742 CWE classification accuracy versus 0.1455 for source (while source leads in Binary F1 at 0.5069 against 0.3116).
  • Shared base and setup: both adapters are trained on synthetic SARD Juliet C programs on top of Qwen2.5-Coder-7B-Instruct under the Apache 2.0 license.

What you can do

  • Reproduce the research experiment: compare how an LLM behaves when analyzing raw source code versus low-level LLVM IR for the exact same functions.
  • Explore vulnerability attribution: test classification across 13 distinct CWE categories using synthetic examples from the SARD Juliet suite.

Example

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
from peft import PeftModel

base_id = "Qwen/Qwen2.5-Coder-7B-Instruct"
repo = "phazertron/vulnerability-detector-ir-source"
bnb = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4", bnb_4bit_compute_dtype=torch.bfloat16)
model = PeftModel.from_pretrained(AutoModelForCausalLM.from_pretrained(base_id, quantization_config=bnb, device_map="auto"), repo, subfolder="ir")
Read moreDetails
Metrics (246 samples): Binary F1 — 0.5069 (source) and 0.3116 (ir); CWE accuracy when detected — 0.1455 (source) and 0.7742 (ir).
QLoRA settings: rank 32, alpha 64, dropout 0.05, 2048-token context budget, 4-bit NF4.
Stack: PEFT 0.13.2, TRL 1.5.0, Transformers 4.57.6.
Limitations: trained on synthetic SARD Juliet data and does not generalize to real production code.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 14 GB · ready files16 GB
  • FP8 / INT8recommendedweights 7.4 GB8.4 GB
  • 4-bitweights 4.3 GB5.3 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 3080 · BF16 — from RTX 5080

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 Qwen2.5-Coder-7B-Instruct.

Language models: with a 8192-token context.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

A ready script for the machine we advise above: paste it and the model downloads and starts.

  1. 1. Rent A30

    $0.26 an hour with the coupon. Pick the Ubuntu image when you create the machine.

    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.

    # phazertron/vulnerability-detector-ir-source — A30, FP8 / INT8
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    # NVIDIA driver: CUDA 13 needs 580+. If older or missing, the current nvidia-open is installed (keep yours: KEEP_DRIVER=1)
    DRV=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -1 | cut -d. -f1)
    case "$DRV" in ""|*[!0-9]*) DRV=0 ;; esac
    if [ "$DRV" -lt 580 ] && [ -z "${KEEP_DRIVER:-}" ]; then
      . /etc/os-release && \
      curl -fsSLO "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv venv ~/vllm --python 3.12 && source ~/vllm/bin/activate
    uv pip install vllm
    vllm serve Qwen/Qwen2.5-Coder-7B-Instruct --max-model-len 8192 --quantization fp8 --enable-lora --max-lora-rank 64 --lora-modules vulnerability-detector-ir-source=phazertron/vulnerability-detector-ir-source
  3. 3. Use it

    The server listens on port 8000 and speaks the OpenAI API: http://localhost:8000/v1 on the machine. From your computer: ssh -L 8000:localhost:8000 user@machine-IP.

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.

  • MTP (multi-token prediction)0.34.0from November 16, 2026new release

    1.5–2× generation speed

    If the model has MTP layers (DeepSeek-V3/R1, Qwen3-Next, GLM, MiMo and others), it predicts several tokens per step itself.

    How to enable: vLLM / SGLang: --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}'. Check the model card.

    Earlier generations (2)
    • EAGLE-3 — Speculative decoding with a small draft head. For models without their own MTP layers.
    • EAGLE-2 / Medusa — Earlier draft-head generations. Still work if that is all you find.
  • FP8 KV cache0.34.0from November 16, 2026new release

    2× more context in the same memory

    Keeps the attention cache in 8 bits, so long contexts fit where they did not.

    How to enable: vLLM: --kv-cache-dtype fp8; llama.cpp: -ctk q8_0 -ctv q8_0.

  • vLLM / SGLang0.34.0from November 16, 2026new release

    many times more tokens/s on a server

    Server engines: batching, FlashAttention, FP8 and speculative decoding out of the box.

    How to enable: vllm serve <model> — an OpenAI-compatible API.

    Earlier generations (1)
    • llama.cpp — GGUF models on any GPU, Mac or even CPU; some layers can stay in RAM.

Model family

Requires the base model

Qwen/Qwen2.5-Coder-7B-Instruct

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

  1. Qwen2.5-7Bfamily root
  2. Qwen2.5-Coder-7B
  3. Qwen2.5-Coder-7B-Instruct
  4. vulnerability-detector-ir-source

New posts about versions of Qwen2.5-7B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 1All on Hugging Face
Merges 5All on Hugging Face

About the model

phazertron/vulnerability-detector-ir-source
Author
phazertron
Category
Language models
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