app.py

import streamlit as st
from transformers import (
    AutoTokenizer,
    AutoModelForSequenceClassification,
    AutoModelForSeq2SeqLM,
)
import torch
import os

# Define the model names and their corresponding Hugging Face models
MODEL_MAPPING = {
    "text2shellcommands": "t5-small",  # Example seq2seq model for generating shell commands
    "pentest_ai": "bert-base-uncased",  # Example classification model for pentesting tasks
}

# Function to create a sidebar for model selection
def select_model():
    """
    Adds a dropdown to the Streamlit sidebar for selecting a model.
    Returns:
        str: The selected model key from MODEL_MAPPING.
    """
    st.sidebar.header("Model Configuration")
    selected_model = st.sidebar.selectbox("Select a model", list(MODEL_MAPPING.keys()))
    return selected_model


# Function to load the model and tokenizer with caching
@st.cache_resource
def load_model_and_tokenizer(model_name):
    """
    Loads the tokenizer and model for the specified Hugging Face model name.
    Uses caching to optimize performance.

    Args:
        model_name (str): The name of the Hugging Face model to load.

    Returns:
        tuple: A tokenizer and model instance.
    """
    try:
        # Load the tokenizer
        tokenizer = AutoTokenizer.from_pretrained(model_name)

        # Determine the correct model class to use
        if "t5" in model_name or "seq2seq" in model_name:
            # Load a sequence-to-sequence model
            model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
        else:
            # Load a sequence classification model
            model = AutoModelForSequenceClassification.from_pretrained(model_name)

        return tokenizer, model
    except Exception as e:
        # Display an error message in the Streamlit app
        st.error(f"An error occurred while loading the model or tokenizer: {str(e)}")
        return None, None


# Function to handle predictions based on the selected model
def predict_with_model(user_input, model, tokenizer, model_choice):
    """
    Handles predictions using the loaded model and tokenizer.

    Args:
        user_input (str): Text input from the user.
        model: Loaded Hugging Face model.
        tokenizer: Loaded Hugging Face tokenizer.
        model_choice (str): Selected model key from MODEL_MAPPING.

    Returns:
        dict: A dictionary containing the prediction results.
    """
    if model_choice == "text2shellcommands":
        # Generate shell commands (Seq2Seq task)
        inputs = tokenizer(user_input, return_tensors="pt", padding=True, truncation=True)
        with torch.no_grad():
            outputs = model.generate(**inputs)
        generated_command = tokenizer.decode(outputs[0], skip_special_tokens=True)
        return {"Generated Shell Command": generated_command}
    else:
        # Perform classification
        inputs = tokenizer(user_input, return_tensors="pt", padding=True, truncation=True)
        with torch.no_grad():
            outputs = model(**inputs)
        logits = outputs.logits
        predicted_class = torch.argmax(logits, dim=-1).item()
        return {
            "Predicted Class": predicted_class,
            "Logits": logits.tolist(),
        }


# Function to process uploaded files
def process_uploaded_file(uploaded_file):
    """
    Reads and processes the uploaded file. Supports text and CSV files.

    Args:
        uploaded_file: The uploaded file.

    Returns:
        str: The content of the file as a string.
    """
    try:
        if uploaded_file is not None:
            file_type = uploaded_file.type

            # Text file processing
            if "text" in file_type:
                content = uploaded_file.read().decode("utf-8")
                return content
            # CSV file processing
            elif "csv" in file_type:
                import pandas as pd
                df = pd.read_csv(uploaded_file)
                return df.to_string()  # Convert the dataframe to string
            else:
                st.error("Unsupported file type. Please upload a text or CSV file.")
                return None
    except Exception as e:
        st.error(f"Error processing file: {e}")
        return None


# Main function to define the Streamlit app
def main():
    st.title("AI Model Inference Dashboard")
    st.markdown(
        """
        This dashboard allows you to interact with different AI models for inference tasks, 
        such as generating shell commands or performing text classification.
        """
    )

    # Model selection
    model_choice = select_model()
    model_name = MODEL_MAPPING.get(model_choice)
    tokenizer, model = load_model_and_tokenizer(model_name)

    # Input text area or file upload
    input_choice = st.radio("Choose Input Method", ("Text Input", "Upload File"))

    if input_choice == "Text Input":
        user_input = st.text_area("Enter your text input:", placeholder="Type your text here...")

        # Handle prediction after submit
        submit_button = st.button("Submit")

        if submit_button and user_input:
            st.write("### Prediction Results:")
            result = predict_with_model(user_input, model, tokenizer, model_choice)
            for key, value in result.items():
                st.write(f"**{key}:** {value}")

    elif input_choice == "Upload File":
        uploaded_file = st.file_uploader("Choose a text or CSV file", type=["txt", "csv"])

        # Handle prediction after submit
        submit_button = st.button("Submit")

        if submit_button and uploaded_file:
            file_content = process_uploaded_file(uploaded_file)
            if file_content:
                st.write("### File Content:")
                st.write(file_content)
                result = predict_with_model(file_content, model, tokenizer, model_choice)
                st.write("### Prediction Results:")
                for key, value in result.items():
                    st.write(f"**{key}:** {value}")
            else:
                st.info("No valid content found in the file.")


if __name__ == "__main__":
    main()