app.py

import time
import streamlit as st
from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.embeddings.base import Embeddings
from langchain.vectorstores import FAISS
from langchain.indexes import VectorstoreIndexCreator
from langchain.chains import RetrievalQA
from langchain.chat_models import ChatOpenAI
from typing import List
from together import Together


st.set_page_config(page_title="چت‌ بات ارتش", 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, 1, 1])
with col2:
    st.image("army.png", width=240)

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


class TogetherEmbeddings(Embeddings):
    def __init__(self, model_name: str, api_key: str):
        self.model_name = model_name
        self.client = Together(api_key=api_key)

    def embed_documents(self, texts: List[str]) -> List[List[float]]:
        response = self.client.embeddings.create(model=self.model_name, input=texts)
        return [item.embedding for item in response.data]

    def embed_query(self, text: str) -> List[float]:
        return self.embed_documents([text])[0]

@st.cache_resource
def get_pdf_index():
    with st.spinner('📄 در حال پردازش فایل PDF...'):
        # لود فایل
        loader = [PyPDFLoader('test1.pdf')]
        pages = []
        for l in loader:
            pages.extend(l.load())

        # اول چانک کردن عادی با سایز 124
        splitter_initial = RecursiveCharacterTextSplitter(
            chunk_size=124, 
            chunk_overlap=25
        )

        small_chunks = []
        for page in pages:
            text = page.page_content
            if len(text) > 124:
                small_chunks.extend(splitter_initial.split_text(text))
            else:
                small_chunks.append(text)

        # حالا چک کنیم هیچ چانکی بیشتر از 3000 کاراکتر نباشه
        final_chunks = []
        final_splitter = RecursiveCharacterTextSplitter(
            chunk_size=2000, 
            chunk_overlap=200
        )

        for chunk in small_chunks:
            if len(chunk) > 2000:
                final_chunks.extend(final_splitter.split_text(chunk))
            else:
                final_chunks.append(chunk)

        # آماده سازی اسناد برای index
        from langchain.schema import Document as LangchainDocument
        documents = [LangchainDocument(page_content=text) for text in final_chunks]

        # ساختن embedding
        embeddings = TogetherEmbeddings(
            model_name="togethercomputer/m2-bert-80M-32k-retrieval",
            api_key="0291f33aee03412a47fa5d8e562e515182dcc5d9aac5a7fb5eefdd1759005979"
        )

        # ساختن ایندکس
        return VectorstoreIndexCreator(
            embedding=embeddings,
            text_splitter=RecursiveCharacterTextSplitter(
                chunk_size=2000, 
                chunk_overlap=200
            )
        ).from_documents(documents)

index = get_pdf_index()

llm = ChatOpenAI(
    base_url="https://api.together.xyz/v1",
    api_key='0291f33aee03412a47fa5d8e562e515182dcc5d9aac5a7fb5eefdd1759005979',
    model="meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8"
)

chain = RetrievalQA.from_chain_type(
    llm=llm,
    chain_type='stuff',
    retriever=index.vectorstore.as_retriever(),
    input_key='question'
)

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

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

prompt = st.chat_input("چطور می‌تونم کمک کنم؟")

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

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

        response = chain.run(f'question:پاسخ را فقط به زبان فارسی جواب بده {st.session_state.pending_prompt}')
        answer = response.split("Helpful Answer:")[-1].strip()
        if not answer:
            answer = "متأسفم، اطلاعات دقیقی در این مورد ندارم."

        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