app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import os

# Model configuration
BASE_MODEL = "mistralai/Mistral-7B-Instruct-v0.3"
ADAPTER_MODEL = "mistral-hackaton-2026/Mistral_SmartContract_Security"

print("Loading model... This may take a few minutes on first run.")

# Load tokenizer from base model
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
tokenizer.pad_token = tokenizer.pad_token or tokenizer.eos_token

# Check if CUDA is available
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Using device: {device}")

# Load base model (with quantization only if GPU available)
if torch.cuda.is_available():
    try:
        print("Loading with 4-bit quantization...")
        from transformers import BitsAndBytesConfig
        bnb_config = BitsAndBytesConfig(
            load_in_4bit=True,
            bnb_4bit_quant_type="nf4",
            bnb_4bit_compute_dtype=torch.float16,
            bnb_4bit_use_double_quant=False,
        )
        base_model = AutoModelForCausalLM.from_pretrained(
            BASE_MODEL,
            quantization_config=bnb_config,
            device_map="auto",
            torch_dtype=torch.float16,
            trust_remote_code=True
        )
    except Exception as e:
        print(f"Quantization failed ({e}), loading in float16...")
        base_model = AutoModelForCausalLM.from_pretrained(
            BASE_MODEL,
            device_map="auto",
            torch_dtype=torch.float16,
            trust_remote_code=True
        )
else:
    print("Loading on CPU (slower but works without GPU)...")
    base_model = AutoModelForCausalLM.from_pretrained(
        BASE_MODEL,
        device_map="auto",
        torch_dtype=torch.float32,
        trust_remote_code=True,
        low_cpu_mem_usage=True
    )

# Load PEFT adapter
print("Loading fine-tuned adapter...")
model = PeftModel.from_pretrained(base_model, ADAPTER_MODEL)
model.eval()  # Set to evaluation mode

print("Model loaded successfully!")

# Sample contracts for demo
SAMPLE_CONTRACTS = {
    "Reentrancy Attack": '''// SPDX-License-Identifier: MIT
pragma solidity ^0.8.0;

contract VulnerableBank {
    mapping(address => uint256) public balances;

    function deposit() public payable {
        balances[msg.sender] += msg.value;
    }

    function withdraw(uint256 amount) public {
        require(balances[msg.sender] >= amount);
        (bool success, ) = msg.sender.call{value: amount}("");
        require(success);
        balances[msg.sender] -= amount;  // VULNERABLE: State updated after external call
    }
}''',
    "Integer Overflow": '''// SPDX-License-Identifier: MIT
pragma solidity 0.7.0;

contract VulnerableToken {
    mapping(address => uint256) public balances;

    function transfer(address to, uint256 amount) public {
        balances[msg.sender] -= amount;  // VULNERABLE: No underflow check
        balances[to] += amount;
    }
}''',
    "Access Control Missing": '''// SPDX-License-Identifier: MIT
pragma solidity ^0.8.0;

contract VulnerableAdmin {
    address public owner;

    constructor() {
        owner = msg.sender;
    }

    function setOwner(address newOwner) public {  // VULNERABLE: Anyone can change owner
        owner = newOwner;
    }
}''',
    "Secure Contract": '''// SPDX-License-Identifier: MIT
pragma solidity ^0.8.0;

contract SecureVault {
    mapping(address => uint256) public balances;
    bool private locked;

    modifier noReentrant() {
        require(!locked, "No reentrancy");
        locked = true;
        _;
        locked = false;
    }

    function withdraw(uint256 amount) public noReentrant {
        require(balances[msg.sender] >= amount);
        balances[msg.sender] -= amount;
        (bool success, ) = msg.sender.call{value: amount}("");
        require(success);
    }
}'''
}

def analyze_contract(contract_code, show_tokens=False):
    """Analyze smart contract for vulnerabilities"""
    
    if not contract_code.strip():
        return "⚠️ Please enter a smart contract to analyze."
    
    # Prepare input
    input_text = f"""<s>[INST] Analyze this contract and provide structured security report:

{contract_code} [/INST]

"""
    
    inputs = tokenizer(input_text, return_tensors="pt", truncation=True, max_length=2048)
    
    if torch.cuda.is_available():
        inputs = inputs.to("cuda")
    
    # Generate analysis
    model.eval()
    with torch.no_grad():
        outputs = model.generate(
            **inputs, 
            max_new_tokens=300, 
            temperature=0.3,
            do_sample=True,
            top_p=0.95
        )
    
    response = tokenizer.decode(outputs[0], skip_special_tokens=False)
    analysis = response.split('[/INST]')[-1].strip()
    
    # Format output
    if show_tokens:
        output = f"### 🔍 Security Analysis (with Custom Tokens)\n\n```\n{analysis}\n```"
    else:
        # Clean up for display
        clean_analysis = analysis.replace('</s>', '').strip()
        
        # Check for vulnerabilities
        is_vulnerable = any([
            '<CRITICAL>' in analysis,
            '<HIGH>' in analysis,
            '<VULN_' in analysis
        ])
        
        if is_vulnerable:
            output = f"""### 🚨 VULNERABILITY DETECTED

{clean_analysis}

---
**Model**: Mistral-7B Fine-tuned with 38 Custom Security Tokens
**Accuracy**: 99.6% | **Precision**: 100% | **Recall**: 99.3%
"""
        else:
            output = f"""### ✅ CONTRACT APPEARS SECURE

{clean_analysis}

---
**Model**: Mistral-7B Fine-tuned with 38 Custom Security Tokens
**Accuracy**: 99.6% | **Precision**: 100% | **Recall**: 99.3%
"""
    
    return output

# Create Gradio interface
with gr.Blocks(theme=gr.themes.Soft(), title="Smart Contract Security Analyzer") as demo:
    gr.Markdown("""
    # 🔐 Smart Contract Security Analyzer
    ### Powered by Fine-Tuned Mistral-7B with Custom Security Tokens
    
    **Hackathon Submission Features:**
    - ✨ 38 custom security tokens (NEW capability)
    - 📊 99.6% accuracy on 30K balanced dataset
    - 🎯 100% precision (zero false positives)
    - 📈 +28.6% improvement over base Mistral
    - 🏗️ Structured XML-style vulnerability reports
    """)
    
    with gr.Row():
        with gr.Column(scale=2):
            contract_input = gr.Code(
                label="📝 Paste Solidity Contract Here",
                language="solidity",
                lines=20,
                placeholder="// SPDX-License-Identifier: MIT\npragma solidity ^0.8.0;\n\ncontract MyContract {\n    // Your contract code here\n}"
            )
            
            with gr.Row():
                analyze_btn = gr.Button("🔍 Analyze Contract", variant="primary", size="lg")
                clear_btn = gr.Button("🗑️ Clear", size="lg")
            
            show_tokens_checkbox = gr.Checkbox(
                label="Show custom security tokens in output",
                value=False
            )
        
        with gr.Column(scale=1):
            gr.Markdown("### 📚 Sample Contracts")
            sample_dropdown = gr.Dropdown(
                choices=list(SAMPLE_CONTRACTS.keys()),
                label="Load Example",
                value=None
            )
            load_sample_btn = gr.Button("Load Sample", size="sm")
    
    output_display = gr.Markdown(label="Analysis Result")
    
    # Performance stats
    gr.Markdown("""
    ---
    ### 📊 Model Performance Metrics
    
    | Metric | Base Mistral | Fine-Tuned | Improvement |
    |--------|-------------|------------|-------------|
    | **Accuracy** | 71.0% | **99.6%** | +28.6% |
    | **Precision** | 64.2% | **100.0%** | +35.8% |
    | **Recall** | 100.0% | 99.3% | -0.7% |
    | **F1 Score** | 0.782 | **0.996** | +0.214 |
    | **Custom Tokens** | 0/38 | **25/38** | NEW! |
    | **Structured Output** | No | **Yes** | NEW! |
    
    **Training Details:**
    - 30,000 balanced samples (50% vulnerable, 50% safe)
    - 6 vulnerability types: Reentrancy, Overflow, Access Control, Unchecked Call, DoS, Timestamp
    - 4-bit quantization with LoRA (1.1% trainable params)
    - 5.5 hours training on Google Colab G4 GPU
    """)
    
    # Event handlers
    def load_sample(sample_name):
        if sample_name:
            return SAMPLE_CONTRACTS[sample_name]
        return ""
    
    analyze_btn.click(
        fn=analyze_contract,
        inputs=[contract_input, show_tokens_checkbox],
        outputs=output_display
    )
    
    load_sample_btn.click(
        fn=load_sample,
        inputs=sample_dropdown,
        outputs=contract_input
    )
    
    clear_btn.click(
        fn=lambda: ("", ""),
        outputs=[contract_input, output_display]
    )

# Launch demo
if __name__ == "__main__":
    demo.launch()