Sabreen2026/codebert-solidity-vulnerability

CodeBERT for Solidity Vulnerability Detection

Fine-tuned version of microsoft/codebert-base for binary classification of Solidity smart contracts as vulnerable (1) or safe (0).

Model Details

• Base model: microsoft/codebert-base
• Task: Binary classification (vulnerable=1, safe=0)
• Dataset: 367 Solidity smart contracts generated by 5 LLMs
• Training: 5 epochs, batch size 8, AdamW optimizer
• Platform: Google Colaboratory (T4 GPU)

Results

• Accuracy: 95.9%
• Precision (vulnerable): 100%
• Recall (vulnerable): 93.3%
• F1-score (vulnerable): 0.97
• False positives: 0

How to Use

Install: pip install transformers torch

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tokenizer = AutoTokenizer.from_pretrained("Sabreen2026/codebert-solidity-vulnerability")
model = AutoModelForSequenceClassification.from_pretrained("Sabreen2026/codebert-solidity-vulnerability")
model.eval()

contract_code = "pragma solidity ^0.8.0; contract Test { }"

inputs = tokenizer(contract_code, truncation=True, padding=True, max_length=512, return_tensors='pt')
with torch.no_grad():
    outputs = model(**inputs)
    prediction = torch.argmax(outputs.logits, dim=1).item()

print('VULNERABLE' if prediction == 1 else 'SAFE')

Citation

Ahmadjee, S. (2026). Vulnerability Detection in LLM-Generated Solidity Smart Contracts Using Fine-Tuned CodeBERT.