aaa
Browse files- PdfChatbot.py +114 -0
PdfChatbot.py
ADDED
@@ -0,0 +1,114 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import time
|
2 |
+
import streamlit as st
|
3 |
+
st.set_page_config(page_title="چت بات ارتش", page_icon="🪖", layout="wide")
|
4 |
+
st.markdown("""
|
5 |
+
<style>
|
6 |
+
.main {
|
7 |
+
background-color: #f4f6f7;
|
8 |
+
}
|
9 |
+
.stChatMessage {
|
10 |
+
background-color: #e8f0fe;
|
11 |
+
border-radius: 12px;
|
12 |
+
padding: 10px;
|
13 |
+
margin-bottom: 10px;
|
14 |
+
}
|
15 |
+
</style>
|
16 |
+
""", unsafe_allow_html=True)
|
17 |
+
|
18 |
+
from langchain.document_loaders import PyPDFLoader
|
19 |
+
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
20 |
+
from langchain.embeddings.base import Embeddings
|
21 |
+
from langchain.vectorstores import FAISS
|
22 |
+
from langchain.indexes import VectorstoreIndexCreator
|
23 |
+
from langchain.chains import RetrievalQA
|
24 |
+
from langchain.chat_models import ChatOpenAI
|
25 |
+
from typing import List
|
26 |
+
from together import Together
|
27 |
+
|
28 |
+
class TogetherEmbeddings(Embeddings):
|
29 |
+
def __init__(self, model_name: str, api_key: str):
|
30 |
+
self.model_name = model_name
|
31 |
+
self.client = Together(api_key=api_key)
|
32 |
+
|
33 |
+
def embed_documents(self, texts: List[str]) -> List[List[float]]:
|
34 |
+
response = self.client.embeddings.create(
|
35 |
+
model=self.model_name,
|
36 |
+
input=texts
|
37 |
+
)
|
38 |
+
return [item.embedding for item in response.data]
|
39 |
+
|
40 |
+
def embed_query(self, text: str) -> List[float]:
|
41 |
+
return self.embed_documents([text])[0]
|
42 |
+
|
43 |
+
@st.cache_resource
|
44 |
+
def get_pdf_index():
|
45 |
+
with st.spinner('لطفاً لحظهای صبر کنید...'):
|
46 |
+
pdf_reader = [PyPDFLoader('C:/Users/ici/Desktop/test1.pdf')]
|
47 |
+
embeddings = TogetherEmbeddings(
|
48 |
+
model_name="togethercomputer/m2-bert-80M-8k-retrieval",
|
49 |
+
api_key="0291f33aee03412a47fa5d8e562e515182dcc5d9aac5a7fb5eefdd1759005979"
|
50 |
+
)
|
51 |
+
return VectorstoreIndexCreator(
|
52 |
+
embedding=embeddings,
|
53 |
+
text_splitter=RecursiveCharacterTextSplitter(chunk_size=300, chunk_overlap=0)
|
54 |
+
).from_loaders(pdf_reader)
|
55 |
+
|
56 |
+
index = get_pdf_index()
|
57 |
+
llm = ChatOpenAI(
|
58 |
+
base_url="https://api.together.xyz/v1",
|
59 |
+
api_key='0291f33aee03412a47fa5d8e562e515182dcc5d9aac5a7fb5eefdd1759005979',
|
60 |
+
model="meta-llama/Llama-3.3-70B-Instruct-Turbo-Free"
|
61 |
+
)
|
62 |
+
chain = RetrievalQA.from_chain_type(
|
63 |
+
llm=llm,
|
64 |
+
chain_type='stuff',
|
65 |
+
retriever=index.vectorstore.as_retriever(),
|
66 |
+
input_key='question'
|
67 |
+
)
|
68 |
+
|
69 |
+
# --- UI زیباسازی ---
|
70 |
+
|
71 |
+
col1, col2 = st.columns([1, 10])
|
72 |
+
with col1:
|
73 |
+
st.image("army.png", width=70)
|
74 |
+
with col2:
|
75 |
+
st.title('🤖 چتبات هوشمند ارتش')
|
76 |
+
|
77 |
+
if 'messages' not in st.session_state:
|
78 |
+
st.session_state.messages = []
|
79 |
+
|
80 |
+
if 'pending_prompt' not in st.session_state:
|
81 |
+
st.session_state.pending_prompt = None
|
82 |
+
|
83 |
+
for message in st.session_state.messages:
|
84 |
+
with st.chat_message(message['role']):
|
85 |
+
st.markdown(f"🗨️ {message['content']}", unsafe_allow_html=True)
|
86 |
+
|
87 |
+
prompt = st.chat_input('چطور میتونم کمک کنم؟')
|
88 |
+
|
89 |
+
if prompt:
|
90 |
+
st.session_state.messages.append({'role': 'user', 'content': prompt})
|
91 |
+
st.session_state.pending_prompt = prompt
|
92 |
+
st.rerun()
|
93 |
+
|
94 |
+
if st.session_state.pending_prompt:
|
95 |
+
with st.chat_message('ai'):
|
96 |
+
thinking_placeholder = st.empty()
|
97 |
+
thinking_placeholder.markdown("🤖 در حال فکر کردن...")
|
98 |
+
|
99 |
+
response = chain.run(f'persian {st.session_state.pending_prompt}')
|
100 |
+
helpful_answer = response.split("Helpful Answer:")[-1]
|
101 |
+
if not helpful_answer.strip():
|
102 |
+
helpful_answer = "اطلاعات دقیقی در دسترس نیست، اما میتوانم به شما کمک کنم تا از منابع دیگر بررسی کنید."
|
103 |
+
|
104 |
+
thinking_placeholder.empty()
|
105 |
+
full_response = ""
|
106 |
+
placeholder = st.empty()
|
107 |
+
for chunk in helpful_answer.split():
|
108 |
+
full_response += chunk + " "
|
109 |
+
placeholder.markdown(full_response + "▌")
|
110 |
+
time.sleep(0.03)
|
111 |
+
|
112 |
+
placeholder.markdown(full_response)
|
113 |
+
st.session_state.messages.append({'role': 'ai', 'content': full_response})
|
114 |
+
st.session_state.pending_prompt = None
|