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
from langchain.chains import RetrievalQA
from langchain_core.retrievers import BaseRetriever
from typing import List
from pydantic import Field
import numpy as np
from sentence_transformers import SentenceTransformer
import faiss
# ----------------- تنظیمات صفحه -----------------
st.set_page_config(page_title="چت بات توانا", page_icon="🪖", layout="wide")
st.markdown("""
""", unsafe_allow_html=True)
col1, col2, col3 = st.columns([1, 2, 1])
with col2:
st.image("army.png", width=240)
st.markdown("""
""", 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')
documents = loader.load()
splitter = RecursiveCharacterTextSplitter(chunk_size=300, chunk_overlap=0)
texts = []
for doc in documents:
texts.extend(splitter.split_text(doc.page_content))
# مدل امبدینگ
model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
# تولید امبدینگها
embeddings = model.encode(texts, convert_to_numpy=True)
# ساخت ایندکس Faiss
index = faiss.IndexFlatL2(embeddings.shape[1])
index.add(embeddings)
docs = [{"text": text} for text in texts]
return docs, embeddings, index, model
# ----------------- تعریف LLM -----------------
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[dict] = Field(...)
embeddings: np.ndarray = Field(...)
index: faiss.Index
model: SentenceTransformer
def _get_relevant_documents(self, query: str) -> List[Document]:
query_embedding = self.model.encode([query], convert_to_numpy=True)
_, indices = self.index.search(query_embedding, 5)
results = []
for i in indices[0]:
results.append(Document(page_content=self.documents[i]['text']))
return results
# ----------------- بارگذاری دیتا -----------------
documents, embeddings, index, model = get_pdf_index()
retriever = SimpleRetriever(
documents=documents,
embeddings=embeddings,
index=index,
model=model
)
# ----------------- ساخت Chain -----------------
qa_chain = RetrievalQA.from_chain_type(
llm=llm,
retriever=retriever,
chain_type="stuff"
)
# ----------------- چت استیت -----------------
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 کاربر -----------------
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()
# ----------------- پاسخدهی مدل -----------------
if st.session_state.pending_prompt:
with st.chat_message('ai'):
thinking = st.empty()
thinking.markdown("🤖 در حال فکر کردن...")
try:
response = qa_chain.run(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