app.py

import os
import subprocess
import streamlit as st
from dotenv import load_dotenv
from xhtml2pdf import pisa
import io
from textwrap import dedent
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

# Load Model Resources
def load_resources():
    load_dotenv()
    huggingface_token = os.getenv("HUGGINGFACE_TOKEN")
    subprocess.run(["huggingface-cli", "login", "--token", huggingface_token], capture_output=True)
    tokenizer = AutoTokenizer.from_pretrained("istiak101/TinyLlama-1.1B-Chat-v0.6-rag-finetunedv2.0")
    model = AutoModelForCausalLM.from_pretrained("istiak101/TinyLlama-1.1B-Chat-v0.6-rag-finetunedv2.0")
    return model, tokenizer

# Chat Prompt
def create_test_prompt(question, context, tokenizer):
    prompt = dedent(
        f"""
    {question}

    Information:

    ```
    {context}
    ```
    """
    )
    messages = [
        {
            "role": "system",
            "content": "Use only the information to answer the question",
        },
        {"role": "user", "content": prompt},
    ]
    return tokenizer.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )

# Store model and tokenizer in session state
if "llama_model" not in st.session_state or "llama_tokenizer" not in st.session_state:
    model, tokenizer = load_resources()
    st.session_state.llama_model = model
    st.session_state.llama_tokenizer = tokenizer

st.set_page_config(page_title="Tell Me Why", page_icon="❓", layout="wide")

def get_llama_response(query):
    # model = st.session_state.llama_model
    # tokenizer = st.session_state.llama_tokenizer

    # inputs = tokenizer(query, return_tensors="pt")
    # outputs = model.generate(**inputs, max_new_tokens=300)
    # response = tokenizer.decode(outputs[0]["generated_text"], skip_special_tokens=True)
    pipe = pipeline(
        task="text-generation",
        model=st.session_state.llama_model,
        tokenizer=st.session_state.llama_tokenizer,
        max_new_tokens=128,
        return_full_text=False,
    )
    outputs = pipe(query)
    return outputs[0]["generated_text"]

# PDF Generation
def generate_pdf(convo, topic):
    html = f"<h2>{topic}</h2><hr>"
    for msg in convo:
        if msg["role"] == "user":
            html += f"<p><strong>You:</strong> {msg['text']}</p>"
        elif msg["role"] == "assistant":
            html += f"<p><strong>AI Assistant:</strong> {msg['text']}</p>"

    result = io.BytesIO()
    pisa_status = pisa.CreatePDF(io.StringIO(html), dest=result)
    if pisa_status.err:
        return None
    return result

# Session Init
if "chat_sessions" not in st.session_state:
    st.session_state.chat_sessions = {}
if "current_conversation" not in st.session_state:
    st.session_state.current_conversation = None
if "edit_mode" not in st.session_state:
    st.session_state.edit_mode = {}

# App Title
st.title("💬 Tell Me Why")

# Custom CSS
st.markdown("""
<style>
.chat-wrapper {
    display: flex;
    flex-direction: column;
}

.message-row {
    display: flex;
    align-items: flex-end;
    justify-content: flex-end;
    margin-top: 10px;
}

.user-bubble {
    background-color: rgba(0, 200, 83, 0.1);
    color: var(--text-color);
    padding: 15px;
    border-radius: 15px;
    max-width: 80%;
    border: 1px solid rgba(0, 200, 83, 0.4);
    position: relative;
    margin-bottom: 20px;
}

.assistant-bubble {
    background-color: rgba(3, 169, 244, 0.1);
    color: var(--text-color);
    padding: 15px;
    border-radius: 15px;
    max-width: 80%;
    align-self: flex-start;
    margin-right: auto;
    border: 1px solid rgba(3, 169, 244, 0.3);
    margin-top: 10px;
    margin-bottom: 40px;
}

.bubble-header {
    font-size: 14px;
    font-weight: bold;
    margin-bottom: 5px;
    display: flex;
    align-items: center;
}

.bubble-header span {
    margin-left: 5px;
}

.icon-col {
    display: flex;
    flex-direction: column;
    gap: 5px;
    align-items: center;
}
</style>
""", unsafe_allow_html=True)

# Sidebar: Conversations
st.sidebar.title("Conversations")
titles = list(st.session_state.chat_sessions.keys())

if titles:
    for topic in titles:
        col1, col2 = st.sidebar.columns([0.8, 0.2])
        if col1.button(f"🎨 {topic}", key=f"select_{topic}"):
            st.session_state.current_conversation = topic
            st.rerun()
        if col2.button("🗑", key=f"delete_{topic}"):
            del st.session_state.chat_sessions[topic]
            if st.session_state.current_conversation == topic:
                st.session_state.current_conversation = None
            st.rerun()
else:
    st.sidebar.write("No conversations yet. Start one below!")

# New Conversation
with st.sidebar.form(key='new_conversation_form', clear_on_submit=True):
    new_topic = st.text_input("New Conversation Name")
    submit_button = st.form_submit_button("Start New Conversation")
    
    if submit_button:
        if new_topic.strip() and new_topic not in st.session_state.chat_sessions:
            st.session_state.chat_sessions[new_topic] = []
            st.session_state.current_conversation = new_topic
            st.sidebar.success(f"Started new conversation: {new_topic}")
            st.rerun()
        elif not new_topic.strip():
            st.sidebar.warning("Please enter a name.")
        else:
            st.sidebar.warning("Conversation already exists!")

# Main Chat Area
if st.session_state.current_conversation:
    convo = st.session_state.chat_sessions[st.session_state.current_conversation]

    st.markdown('<div class="chat-wrapper">', unsafe_allow_html=True)

    for idx, msg in enumerate(convo):
        with st.container():
            if msg["role"] == "user":
                if st.session_state.edit_mode.get(idx, False):
                    # Split the message into question and context
                    question_input, context_input = msg["text"].split("<br><br>")
                    # Remove the "Question:" and "Context:" parts from the beginning
                    question_input = question_input.replace("Question: ", "")
                    context_input = context_input.replace("Context: ", "")

                    # Provide separate inputs for the question and context
                    new_question = st.text_input("Edit your question:", value=question_input, key=f"edit_question_{idx}")
                    new_context = st.text_area("Edit your context:", value=context_input, key=f"edit_context_{idx}")

                    prompt = create_test_prompt(new_question, new_context, st.session_state.llama_tokenizer)
                    col1, col2 = st.columns([1, 1])
                    with col1:
                        if st.button("✅ Save", key=f"save_{idx}"):
                            # Combine question and context without the "Question:" and "Context:" labels
                            new_combined_input = f"{new_question}<br><br>{new_context}"
                            msg["text"] = new_combined_input
                            with st.spinner("Generating response..."):
                                try:
                                    new_response = get_llama_response(prompt)
                                except:
                                    new_response = "Failed to retrieve response."
                            if idx + 1 < len(convo) and convo[idx + 1]["role"] == "assistant":
                                convo[idx + 1]["text"] = new_response
                            st.session_state.edit_mode[idx] = False
                            st.session_state.chat_sessions[st.session_state.current_conversation] = convo
                            st.rerun()
                    with col2:
                        if st.button("❌ Cancel", key=f"cancel_{idx}"):
                            st.session_state.edit_mode[idx] = False
                            st.rerun()
                else:
                    col1, col2 = st.columns([0.1, 0.9])
                    with col1:
                        if st.button("✏️", key=f"edit_btn_{idx}"):
                            st.session_state.edit_mode[idx] = True
                            st.rerun()
                    with col2:
                        st.markdown(f'''
                        <div class="user-bubble">
                            <div class="bubble-header">👤 <span>You</span></div>
                            {msg["text"]}
                        </div>
                        ''', unsafe_allow_html=True)

            elif msg["role"] == "assistant":
                st.markdown(f'''
                <div class="assistant-bubble">
                    <div class="bubble-header">📘 <span>AI Assistant</span></div>
                    {msg["text"]}
                </div>
                ''', unsafe_allow_html=True)

    st.markdown('</div>', unsafe_allow_html=True)

    # User Prompt
    if len(convo) % 2 == 1:
        last_user_msg = convo[-1]["text"]
        question_input, context_input = last_user_msg.split("<br><br>")
        question_input = question_input.replace("Question: ", "")
        context_input = context_input.replace("Context: ", "")
    
        prompt = create_test_prompt(question_input, context_input, st.session_state.llama_tokenizer)
        with st.spinner("Generating response..."):
            try:
                assistant_reply = get_llama_response(prompt)
                
            except Exception as e:
                assistant_reply = f"⚠️ Failed to generate response"
        convo.append({"role": "assistant", "text": assistant_reply})
        st.session_state.chat_sessions[st.session_state.current_conversation] = convo
        st.rerun()

    st.markdown("---")
    # Export PDF
    if st.button("📥 Export Conversation as PDF"):
        pdf_bytes = generate_pdf(convo, st.session_state.current_conversation)
        if pdf_bytes:
            st.download_button("Download PDF", pdf_bytes, file_name="TellMeWhy_Conversation.pdf", mime="application/pdf")
        else:
            st.error("❌ Failed to generate PDF.")
            
    with st.form(key="submit_form", clear_on_submit=True):
        question_input = st.text_input("Enter your question:")
        context_input = st.text_area("Enter your context:")
        submit_button = st.form_submit_button("Submit")
        if submit_button:
            if question_input and context_input:
                combined_input = f"Question: {question_input}<br><br>Context: {context_input}"
                convo.append({"role": "user", "text": combined_input})
                st.session_state.chat_sessions[st.session_state.current_conversation] = convo
                st.rerun()