app.py

# app.py

import os
import gradio as gr

# Import your modules
from parsers import parse_pcap
from analysis import analyze_calls
from call_flow import create_call_flow_diagram

# We'll adapt llm_utils to load flan-t5-base locally
from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM

def create_local_pipeline(model_id="google/flan-t5-base"):
    """
    Create a local pipeline for Flan-T5, which is a seq2seq model.
    This should run within ~16GB RAM for 'flan-t5-base'.
    """
    tokenizer = AutoTokenizer.from_pretrained(model_id)
    model = AutoModelForSeq2SeqLM.from_pretrained(model_id, device_map="auto")
    # Use text2text-generation for T5-based models
    return pipeline("text2text-generation", model=model, tokenizer=tokenizer)

# Initialize model pipeline once
MODEL_ID = "google/flan-t5-base"
generator = create_local_pipeline(MODEL_ID)

def get_llm_opinion(prompt, generator):
    """
    Generate text from the local Flan-T5 pipeline.
    We use text2text-generation to handle the seq2seq nature of T5.
    """
    outputs = generator(prompt, max_length=256, do_sample=True, temperature=0.5)
    return outputs[0]["generated_text"]

def process_file(pcap_file):
    """
    This function is called when user clicks the 'Analyze File' button.
    - pcap_file: A dictionary returned by Gradio's File component.
    Returns:
    - a textual summary (analysis + call flow)
    - the dictionary of calls_by_id (saved in Gradio state so we can pass them to the LLM)
    """
    print("DEBUG pcap_file =", pcap_file, type(pcap_file))  # <--- debug line
    if not pcap_file:
        return "No file uploaded.", {}

    # pcap_file is a string path to the temp file, e.g. "/tmp/gradio/xyz/somefile.pcap"
    temp_filename = pcap_file

    # 1) Parse the PCAP
    calls_by_id = parse_pcap(temp_filename)

    # 2) Analyze the calls
    analysis_result = analyze_calls(calls_by_id)

    # 3) Create a call flow diagram (textual)
    call_flow_text = create_call_flow_diagram(calls_by_id)

    # Combine them into one display string
    result_text = (
        f"=== VoIP Analysis ===\n"
        f"{analysis_result}\n\n"
        f"=== Call Flow Diagram(s) ===\n"
        f"{call_flow_text}"
    )

    # Delete the temp file if desired
    os.remove(temp_filename)

    return result_text, calls_by_id

def ask_llm_opinion(calls_data, question):
    """
    This function passes the call analysis info + user question to the LLM (local pipeline).
    - calls_data: The dictionary of calls returned from parse_pcap() [Gradio state]
    - question: The user’s question in text form
    """
    if not calls_data:
        return "No call data available. Please upload and analyze a PCAP first."

    if not question.strip():
        return "Please enter a question."

    # Summarize the calls for context
    calls_context = "Below is a brief description of the calls found in the PCAP:\n"
    for call_id, call_obj in calls_data.items():
        calls_context += f"- Call-ID: {call_id}, from_tag: {call_obj.from_tag}, to_tag: {call_obj.to_tag}\n"

    # Build a prompt for T5
    prompt = (
        f"{calls_context}\n"
        f"User's question: {question}\n\n"
        "Please provide your expert VoIP analysis or advice."
    )

    # Query the local pipeline
    llm_response = get_llm_opinion(prompt, generator=generator)
    return llm_response

def main():
    """
    Build the Gradio interface with two tabs:
    1) PCAP Analysis
    2) LLM Consultation
    """
    with gr.Blocks() as demo:
        gr.Markdown("# VoIP Analyzer\nUpload a PCAP/PCAPNG file for SIP/RTP analysis. Then consult Flan-T5 for further insights.")

        # We keep the calls data in a Gradio State so we can pass it between tabs
        calls_state = gr.State({})

        with gr.Tab("PCAP Analysis"):
            file_input = gr.File(label="Upload a PCAP or PCAPNG file")
            analyze_button = gr.Button("Analyze File")
            analysis_output = gr.Textbox(label="Analysis & Call Flow", lines=20)

            analyze_button.click(
                fn=process_file,
                inputs=file_input,
                outputs=[analysis_output, calls_state]
            )

        with gr.Tab("LLM Consultation"):
            question_input = gr.Textbox(label="Ask a question about the call(s)")
            ask_button = gr.Button("Ask LLM")
            llm_output = gr.Textbox(label="LLM Response", lines=10)

            ask_button.click(
                fn=ask_llm_opinion,
                inputs=[calls_state, question_input],
                outputs=[llm_output]
            )

    demo.launch(server_name="0.0.0.0", server_port=7860, share=True)

if __name__ == "__main__":
    main()