app.py

import os
import time
import streamlit as st
from langchain.chat_models import ChatOpenAI
from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.schema import Document as LangchainDocument
from langchain.chains import RetrievalQA
import torch
from langchain_core.retrievers import BaseRetriever
from langchain_core.documents import Document
from typing import List
from pydantic import Field
from sentence_transformers import SentenceTransformer
import numpy as np
from langchain.indexes.vectorstore import VectorstoreIndexCreator

from sentence_transformers import SentenceTransformer
import faiss


# ----------------- تنظیمات صفحه -----------------
st.set_page_config(page_title="چت‌ بات توانا", page_icon="🪖", layout="wide")

st.markdown("""
    <style>
    @import url('https://fonts.googleapis.com/css2?family=Vazirmatn:wght@400;700&display=swap');
    html, body, [class*="css"] {
        font-family: 'Vazirmatn', Tahoma, sans-serif;
        direction: rtl;
        text-align: right;
    }
    .stApp {
        background: url("./military_bg.jpeg") no-repeat center center fixed;
        background-size: cover;
        backdrop-filter: blur(2px);
    }
    .stChatMessage {
        background-color: rgba(255,255,255,0.8);
        border: 1px solid #4e8a3e;
        border-radius: 12px;
        padding: 16px;
        margin-bottom: 15px;
        box-shadow: 0 4px 10px rgba(0,0,0,0.2);
        animation: fadeIn 0.4s ease-in-out;
    }
    .stTextInput > div > input, .stTextArea textarea {
        background-color: rgba(255,255,255,0.9) !important;
        border-radius: 8px !important;
        direction: rtl;
        text-align: right;
        font-family: 'Vazirmatn', Tahoma;
    }
    .stButton>button {
        background-color: #4e8a3e !important;
        color: white !important;
        font-weight: bold;
        border-radius: 10px;
        padding: 8px 20px;
        transition: 0.3s;
    }
    .stButton>button:hover {
        background-color: #3c6d30 !important;
    }
    .header-text {
        text-align: center;
        margin-top: 20px;
        margin-bottom: 40px;
        background-color: rgba(255, 255, 255, 0.75);
        padding: 20px;
        border-radius: 20px;
        box-shadow: 0 4px 12px rgba(0,0,0,0.2);
    }
    .header-text h1 {
        font-size: 42px;
        color: #2c3e50;
        margin: 0;
        font-weight: bold;
    }
    .subtitle {
        font-size: 18px;
        color: #34495e;
        margin-top: 8px;
    }
    @keyframes fadeIn {
        from { opacity: 0; transform: translateY(10px); }
        to { opacity: 1; transform: translateY(0); }
    }
    </style>
""", unsafe_allow_html=True)

col1, col2, col3 = st.columns([1, 2, 1])
with col2:
    st.image("army.png", width=240)

st.markdown("""
    <div class="header-text">
        <h1>چت‌ بات توانا</h1>
        <div class="subtitle">دستیار هوشمند برای تصمیم‌گیری در میدان نبرد</div>
    </div>
""", unsafe_allow_html=True)

# ----------------- لود PDF و ساخت ایندکس -----------------

@st.cache_resource
def get_pdf_index():
    with st.spinner('📄 در حال پردازش فایل PDF...'):
        loader = [PyPDFLoader('test1.pdf')]

        model_name = "togethercomputer/m2-bert-80M-8k-retrieval"  
        model = SentenceTransformer(model_name, trust_remote_code=True)  

        splitter = RecursiveCharacterTextSplitter(chunk_size=300, chunk_overlap=0)
        texts = []
        for doc in loader:
            texts.extend(splitter.split_text(doc.page_content))

        progress_bar = st.progress(0)
        total_docs = len(texts)

        embeddings = []
        batch_size = 128
        for i in range(0, total_docs, batch_size):
            batch_texts = texts[i:i+batch_size]
            batch_embeddings = model.encode(batch_texts, convert_to_numpy=True)
            embeddings.extend(batch_embeddings)

            progress_bar.progress(min((i + batch_size) / total_docs, 1.0))

        time.sleep(1)
        progress_bar.empty()

        embeddings = np.array(embeddings)
        index = faiss.IndexFlatL2(embeddings.shape[1]) 
        index.add(embeddings)

        return VectorstoreIndexCreator(
            embedding=model.encode,
            text_splitter=splitter
        ).from_loaders(loader)


# ----------------- تعریف LLM از Groq -----------------
llm = ChatOpenAI(
    base_url="https://api.together.xyz/v1",
    api_key='0291f33aee03412a47fa5d8e562e515182dcc5d9aac5a7fb5eefdd1759005979',
    model="meta-llama/Llama-3.3-70B-Instruct-Turbo-Free"
)


# ----------------- تعریف SimpleRetriever -----------------
class SimpleRetriever(BaseRetriever):
    documents: List[Document] = Field(...)
    embeddings: List[np.ndarray] = Field(...)
    index: faiss.Index

    def _get_relevant_documents(self, query: str) -> List[Document]:
        sentence_model = SentenceTransformer("togethercomputer/m2-bert-80M-8k-retrieval", trust_remote_code=True)
        query_embedding = sentence_model.encode(query, convert_to_numpy=True)

        # جستجوی اسناد مشابه
        _, indices = self.index.search(np.expand_dims(query_embedding, axis=0), 5)  # پیدا کردن 5 سند مشابه

        return [self.documents[i] for i in indices[0]]


# ----------------- ساخت Index -----------------
documents, embeddings, index = get_pdf_index()
retriever = SimpleRetriever(documents=documents, embeddings=embeddings, index=index)

# ----------------- ساخت Chain -----------------
chain = RetrievalQA.from_chain_type(
    llm=llm,
    retriever=retriever,
    chain_type="stuff",
    input_key="question"
)

# ----------------- استیت برای چت -----------------
if 'messages' not in st.session_state:
    st.session_state.messages = []

if 'pending_prompt' not in st.session_state:
    st.session_state.pending_prompt = None

# ----------------- نمایش پیام‌های قبلی -----------------
for msg in st.session_state.messages:
    with st.chat_message(msg['role']):
        st.markdown(f"🗨️ {msg['content']}", unsafe_allow_html=True)

# ----------------- ورودی چت -----------------
prompt = st.chat_input("سوالی در مورد فایل بپرس...")

if prompt:
    st.session_state.messages.append({'role': 'user', 'content': prompt})
    st.session_state.pending_prompt = prompt
    st.rerun()

# ----------------- پاسخ مدل -----------------
if st.session_state.pending_prompt:
    with st.chat_message('ai'):
        thinking = st.empty()
        thinking.markdown("🤖 در حال فکر کردن  ...")

        try:
            response = chain.run(f"سوال: {st.session_state.pending_prompt}")
            answer = response.strip()
        except Exception as e:
            answer = f"خطا در پاسخ‌دهی: {str(e)}"

        thinking.empty()

        full_response = ""
        placeholder = st.empty()
        for word in answer.split():
            full_response += word + " "
            placeholder.markdown(full_response + "▌")
            time.sleep(0.03)

        placeholder.markdown(full_response)
        st.session_state.messages.append({'role': 'ai', 'content': full_response})
        st.session_state.pending_prompt = None