Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,21 +1,19 @@
|
|
1 |
import streamlit as st
|
2 |
import os
|
3 |
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
|
4 |
-
|
5 |
-
# --- LANGCHAIN IMPORTS ---
|
6 |
from langchain_community.document_loaders import PyPDFLoader
|
7 |
from langchain_experimental.text_splitter import SemanticChunker
|
8 |
from langchain_huggingface import HuggingFaceEmbeddings
|
9 |
from langchain_community.vectorstores import FAISS
|
10 |
from langchain.memory import ConversationBufferMemory
|
11 |
|
12 |
-
# 1)
|
13 |
st.title("💬 المحادثة التفاعلية - إدارة البيانات وحماية البيانات الشخصية")
|
14 |
local_file = "Policies001.pdf"
|
15 |
|
16 |
index_folder = "faiss_index"
|
17 |
|
18 |
-
#
|
19 |
st.markdown(
|
20 |
"""
|
21 |
<style>
|
@@ -28,15 +26,17 @@ st.markdown(
|
|
28 |
unsafe_allow_html=True
|
29 |
)
|
30 |
|
31 |
-
# 2)
|
32 |
embeddings = HuggingFaceEmbeddings(
|
33 |
model_name="CAMeL-Lab/bert-base-arabic-camelbert-mix",
|
34 |
model_kwargs={"trust_remote_code": True}
|
35 |
)
|
36 |
|
37 |
if os.path.exists(index_folder):
|
|
|
38 |
vectorstore = FAISS.load_local(index_folder, embeddings, allow_dangerous_deserialization=True)
|
39 |
else:
|
|
|
40 |
loader = PyPDFLoader(local_file)
|
41 |
documents = loader.load()
|
42 |
|
@@ -47,68 +47,65 @@ else:
|
|
47 |
)
|
48 |
chunked_docs = text_splitter.split_documents(documents)
|
49 |
|
|
|
50 |
vectorstore = FAISS.from_documents(chunked_docs, embeddings)
|
51 |
vectorstore.save_local(index_folder)
|
52 |
|
53 |
-
# 3)
|
54 |
retriever = vectorstore.as_retriever(search_type="similarity", search_kwargs={"k": 5})
|
55 |
|
56 |
-
# 4)
|
57 |
-
|
58 |
-
model_name = "CohereForAI/c4ai-command-r7b-arabic-02-2025" # Replace with the actual Hugging Face model ID
|
59 |
-
|
60 |
-
# Set Hugging Face token securely
|
61 |
-
hf_token = os.getenv("HF_TOKEN") # Ensure you set your token as an environment variable in Hugging Face Spaces
|
62 |
|
|
|
|
|
63 |
if hf_token is None:
|
64 |
st.error("Hugging Face token not found. Please set the 'HF_TOKEN' environment variable.")
|
65 |
st.stop()
|
66 |
|
67 |
-
#
|
68 |
tokenizer = AutoTokenizer.from_pretrained(model_name, use_auth_token=hf_token)
|
69 |
model = AutoModelForCausalLM.from_pretrained(model_name, use_auth_token=hf_token)
|
70 |
|
71 |
-
#
|
72 |
qa_pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer, device=0)
|
73 |
|
74 |
-
#
|
75 |
memory = ConversationBufferMemory(
|
76 |
-
memory_key="chat_history",
|
77 |
-
return_messages=True
|
78 |
)
|
79 |
|
80 |
-
#
|
81 |
if "messages" not in st.session_state:
|
82 |
st.session_state["messages"] = [
|
83 |
{"role": "assistant", "content": "👋 مرحبًا! اسألني أي شيء عن إدارة البيانات وحماية البيانات الشخصية!"}
|
84 |
]
|
85 |
|
86 |
-
#
|
87 |
for msg in st.session_state["messages"]:
|
88 |
with st.chat_message(msg["role"]):
|
89 |
-
# Apply the "rtl" class to style Arabic text correctly
|
90 |
st.markdown(f'<div class="rtl">{msg["content"]}</div>', unsafe_allow_html=True)
|
91 |
|
92 |
-
#
|
93 |
user_input = st.chat_input("اكتب سؤالك هنا")
|
94 |
|
95 |
-
#
|
96 |
if user_input:
|
97 |
-
#
|
98 |
st.session_state["messages"].append({"role": "user", "content": user_input})
|
99 |
with st.chat_message("user"):
|
100 |
st.markdown(f'<div class="rtl">{user_input}</div>', unsafe_allow_html=True)
|
101 |
|
102 |
-
#
|
103 |
-
# Combine retriever results and user input for context-aware answering
|
104 |
retrieved_docs = retriever.get_relevant_documents(user_input)
|
105 |
context = "\n".join([doc.page_content for doc in retrieved_docs])
|
106 |
full_input = f"السياق:\n{context}\n\nالسؤال:\n{user_input}"
|
107 |
|
108 |
-
#
|
109 |
response = qa_pipeline(full_input, max_length=500, num_return_sequences=1)[0]["generated_text"]
|
110 |
|
111 |
-
#
|
112 |
st.session_state["messages"].append({"role": "assistant", "content": response})
|
113 |
with st.chat_message("assistant"):
|
114 |
st.markdown(f'<div class="rtl">{response}</div>', unsafe_allow_html=True)
|
|
|
1 |
import streamlit as st
|
2 |
import os
|
3 |
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
|
|
|
|
|
4 |
from langchain_community.document_loaders import PyPDFLoader
|
5 |
from langchain_experimental.text_splitter import SemanticChunker
|
6 |
from langchain_huggingface import HuggingFaceEmbeddings
|
7 |
from langchain_community.vectorstores import FAISS
|
8 |
from langchain.memory import ConversationBufferMemory
|
9 |
|
10 |
+
# --- 1) إعداد الصفحة ---
|
11 |
st.title("💬 المحادثة التفاعلية - إدارة البيانات وحماية البيانات الشخصية")
|
12 |
local_file = "Policies001.pdf"
|
13 |
|
14 |
index_folder = "faiss_index"
|
15 |
|
16 |
+
# إضافة CSS مخصص لدعم النصوص من اليمين لليسار
|
17 |
st.markdown(
|
18 |
"""
|
19 |
<style>
|
|
|
26 |
unsafe_allow_html=True
|
27 |
)
|
28 |
|
29 |
+
# --- 2) تحميل أو بناء قاعدة بيانات FAISS ---
|
30 |
embeddings = HuggingFaceEmbeddings(
|
31 |
model_name="CAMeL-Lab/bert-base-arabic-camelbert-mix",
|
32 |
model_kwargs={"trust_remote_code": True}
|
33 |
)
|
34 |
|
35 |
if os.path.exists(index_folder):
|
36 |
+
# تحميل قاعدة البيانات إذا كانت موجودة
|
37 |
vectorstore = FAISS.load_local(index_folder, embeddings, allow_dangerous_deserialization=True)
|
38 |
else:
|
39 |
+
# تحميل PDF وتقسيم النصوص
|
40 |
loader = PyPDFLoader(local_file)
|
41 |
documents = loader.load()
|
42 |
|
|
|
47 |
)
|
48 |
chunked_docs = text_splitter.split_documents(documents)
|
49 |
|
50 |
+
# إنشاء قاعدة بيانات FAISS
|
51 |
vectorstore = FAISS.from_documents(chunked_docs, embeddings)
|
52 |
vectorstore.save_local(index_folder)
|
53 |
|
54 |
+
# --- 3) إعداد المسترجع ---
|
55 |
retriever = vectorstore.as_retriever(search_type="similarity", search_kwargs={"k": 5})
|
56 |
|
57 |
+
# --- 4) إعداد نموذج النص ---
|
58 |
+
model_name = "CohereForAI/c4ai-command-r7b-arabic-02-2025" # اسم النموذج
|
|
|
|
|
|
|
|
|
59 |
|
60 |
+
# التأكد من وجود توكن Hugging Face
|
61 |
+
hf_token = os.getenv("HF_TOKEN")
|
62 |
if hf_token is None:
|
63 |
st.error("Hugging Face token not found. Please set the 'HF_TOKEN' environment variable.")
|
64 |
st.stop()
|
65 |
|
66 |
+
# تحميل النموذج والمحول
|
67 |
tokenizer = AutoTokenizer.from_pretrained(model_name, use_auth_token=hf_token)
|
68 |
model = AutoModelForCausalLM.from_pretrained(model_name, use_auth_token=hf_token)
|
69 |
|
70 |
+
# إعداد pipeline لتوليد النصوص
|
71 |
qa_pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer, device=0)
|
72 |
|
73 |
+
# --- 5) إعداد الذاكرة ---
|
74 |
memory = ConversationBufferMemory(
|
75 |
+
memory_key="chat_history",
|
76 |
+
return_messages=True
|
77 |
)
|
78 |
|
79 |
+
# --- 6) إدارة رسائل المستخدم ---
|
80 |
if "messages" not in st.session_state:
|
81 |
st.session_state["messages"] = [
|
82 |
{"role": "assistant", "content": "👋 مرحبًا! اسألني أي شيء عن إدارة البيانات وحماية البيانات الشخصية!"}
|
83 |
]
|
84 |
|
85 |
+
# عرض الرسائل الحالية
|
86 |
for msg in st.session_state["messages"]:
|
87 |
with st.chat_message(msg["role"]):
|
|
|
88 |
st.markdown(f'<div class="rtl">{msg["content"]}</div>', unsafe_allow_html=True)
|
89 |
|
90 |
+
# --- 7) إدخال المستخدم ---
|
91 |
user_input = st.chat_input("اكتب سؤالك هنا")
|
92 |
|
93 |
+
# --- 8) معالجة رسالة المستخدم ---
|
94 |
if user_input:
|
95 |
+
# عرض رسالة المستخدم
|
96 |
st.session_state["messages"].append({"role": "user", "content": user_input})
|
97 |
with st.chat_message("user"):
|
98 |
st.markdown(f'<div class="rtl">{user_input}</div>', unsafe_allow_html=True)
|
99 |
|
100 |
+
# استرجاع المستندات ذات الصلة
|
|
|
101 |
retrieved_docs = retriever.get_relevant_documents(user_input)
|
102 |
context = "\n".join([doc.page_content for doc in retrieved_docs])
|
103 |
full_input = f"السياق:\n{context}\n\nالسؤال:\n{user_input}"
|
104 |
|
105 |
+
# توليد الإجابة باستخدام النموذج
|
106 |
response = qa_pipeline(full_input, max_length=500, num_return_sequences=1)[0]["generated_text"]
|
107 |
|
108 |
+
# عرض الإجابة
|
109 |
st.session_state["messages"].append({"role": "assistant", "content": response})
|
110 |
with st.chat_message("assistant"):
|
111 |
st.markdown(f'<div class="rtl">{response}</div>', unsafe_allow_html=True)
|