CVEParrot 🦜

CVEParrot is a Google T5 model fine-tuned on CVE (Common Vulnerabilities and Exposures) database to understand and generate security vulnerability information.

Model Description

• Developed by: Subhay Roy Chowdhury(findthehead)
• Base Model: Google T5 Small

Use Cases

• Generate CVE descriptions
• Analyze vulnerability information
• Security research and analysis
• Automated vulnerability documentation
• CVE information extraction and summarization

Inference Code

import warnings
import os
os.environ["TF_ENABLE_ONEDNN_OPTS"] = "0"
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"
os.environ["TRANSFORMERS_NO_ADVISORY_WARNINGS"] = "1"
warnings.filterwarnings("ignore")
import torch
from transformers import AutoTokenizer, T5ForConditionalGeneration
model_name = "Prachir-AI/cveparrot"
tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=False)


model = T5ForConditionalGeneration.from_pretrained(model_name)

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

prompt = "Provide detailed information about CVE-2021-3184."
inputs = tokenizer(prompt, return_tensors="pt").to(device)

with torch.no_grad():
    output_ids = model.generate(
        **inputs,
        max_new_tokens=128,
        temperature=1.0,
        do_sample=True,
    )

response = tokenizer.decode(output_ids[0], skip_special_tokens=True)
print(response)

Option 2: Using GGUF Model with Ollama (Local Inference)

The model is available in GGUF format for efficient local inference using Ollama.

Note: T5 architecture support in Ollama may be experimental. If you encounter issues, please use the Hugging Face Transformers method (Option 1) or try alternative GGUF inference tools like llama.cpp.

Step 1: Install Ollama

# Linux
curl -fsSL https://ollama.com/install.sh | sh

# macOS
brew install ollama

# Or download from https://ollama.com

Step 2: Download the GGUF Model

Download cveparrot.gguf from the Files section of this repository.

Step 3: Create a Modelfile

Create a file named Modelfile in the same directory as the downloaded GGUF:

FROM ./cveparrot.gguf

TEMPLATE """{{ .Prompt }}"""

PARAMETER temperature 0.7
PARAMETER top_p 0.9
PARAMETER top_k 40
PARAMETER num_ctx 2048

Step 4: Create and Run the Model

# Create the model in Ollama
ollama create cveparrot -f Modelfile

# Interactive mode
ollama run cveparrot

# Single query
ollama run cveparrot "Describe CVE-2024-1234"

Using Ollama API (Python):

pip install ollama

import ollama

# Generate response
response = ollama.generate(
    model='cveparrot',  # Use the local model name you created
    prompt='Describe the security vulnerability CVE-2024-1234',
)

print(response['response'])

Using Ollama API (curl):

curl http://localhost:11434/api/generate -d '{
  "model": "cveparrot",
  "prompt": "Describe CVE-2024-1234",
  "stream": false
}'

Model Files

• model.safetensors: PyTorch model weights in Safetensors format
• cveparrot.gguf: Quantized GGUF model for efficient inference
• tokenizer_config.json: Tokenizer configuration
• config.json: Model configuration
• spiece.model: SentencePiece tokenizer model

Training Details

This model was fine-tuned on CVE database entries to understand and generate security vulnerability information. The training focused on:

• CVE descriptions and technical details
• Vulnerability severity and impact analysis
• Security patches and mitigation strategies
• Affected software and version information