Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,16 +1,18 @@
|
|
1 |
-
import streamlit as st
|
2 |
import os
|
3 |
-
from
|
|
|
|
|
|
|
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
|
@@ -27,56 +29,86 @@ st.markdown(
|
|
27 |
)
|
28 |
|
29 |
# --- 2) تحميل أو بناء قاعدة بيانات FAISS ---
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
|
|
34 |
|
35 |
-
if os.path.exists(index_folder):
|
36 |
-
|
37 |
-
|
38 |
-
else:
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
47 |
)
|
48 |
-
chunked_docs = text_splitter.split_documents(documents)
|
49 |
|
50 |
-
#
|
51 |
-
|
52 |
-
|
|
|
|
|
|
|
|
|
|
|
53 |
|
54 |
-
|
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 |
-
#
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
|
|
|
|
65 |
|
66 |
-
#
|
67 |
-
|
68 |
-
|
|
|
|
|
69 |
|
70 |
-
|
71 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
72 |
|
73 |
-
# --- 5) إعداد الذاكرة ---
|
74 |
-
memory = ConversationBufferMemory(
|
75 |
-
memory_key="chat_history",
|
76 |
-
return_messages=True
|
77 |
-
)
|
78 |
|
79 |
-
# ---
|
|
|
|
|
|
|
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!"}
|
@@ -87,25 +119,20 @@ 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 |
-
#
|
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 |
-
#
|
101 |
-
|
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 |
-
|
|
|
110 |
with st.chat_message("assistant"):
|
111 |
-
st.markdown(f'<div class="rtl">{
|
|
|
|
|
1 |
import os
|
2 |
+
from huggingface_hub import hf_hub_download
|
3 |
+
from langchain.llms import LlamaCpp
|
4 |
+
from langchain.chains import ConversationalRetrievalChain
|
5 |
+
from langchain.memory import ConversationBufferMemory
|
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 |
|
11 |
# --- 1) إعداد الصفحة ---
|
12 |
+
import streamlit as st
|
13 |
+
|
14 |
st.title("💬 المحادثة التفاعلية - إدارة البيانات وحماية البيانات الشخصية")
|
15 |
local_file = "Policies001.pdf"
|
|
|
16 |
index_folder = "faiss_index"
|
17 |
|
18 |
# إضافة CSS مخصص لدعم النصوص من اليمين لليسار
|
|
|
29 |
)
|
30 |
|
31 |
# --- 2) تحميل أو بناء قاعدة بيانات FAISS ---
|
32 |
+
def build_vectorstore():
|
33 |
+
embeddings = HuggingFaceEmbeddings(
|
34 |
+
model_name="CAMeL-Lab/bert-base-arabic-camelbert-mix",
|
35 |
+
model_kwargs={"trust_remote_code": True}
|
36 |
+
)
|
37 |
|
38 |
+
if os.path.exists(index_folder):
|
39 |
+
# تحميل قاعدة البيانات إذا كانت موجودة
|
40 |
+
return FAISS.load_local(index_folder, embeddings, allow_dangerous_deserialization=True)
|
41 |
+
else:
|
42 |
+
# تحميل PDF وتقسيم النصوص
|
43 |
+
loader = PyPDFLoader(local_file)
|
44 |
+
documents = loader.load()
|
45 |
+
|
46 |
+
text_splitter = SemanticChunker(
|
47 |
+
embeddings=embeddings,
|
48 |
+
breakpoint_threshold_type='percentile',
|
49 |
+
breakpoint_threshold_amount=90
|
50 |
+
)
|
51 |
+
chunked_docs = text_splitter.split_documents(documents)
|
52 |
+
|
53 |
+
# إنشاء قاعدة بيانات FAISS
|
54 |
+
vectorstore = FAISS.from_documents(chunked_docs, embeddings)
|
55 |
+
vectorstore.save_local(index_folder)
|
56 |
+
return vectorstore
|
57 |
+
|
58 |
+
|
59 |
+
# --- 3) تحميل النموذج ---
|
60 |
+
def load_llm():
|
61 |
+
"""
|
62 |
+
Downloads a Q4_K_M GGUF model and loads it via llama-cpp.
|
63 |
+
"""
|
64 |
+
# 1) Download the GGUF model from Hugging Face
|
65 |
+
model_file = hf_hub_download(
|
66 |
+
repo_id="DevQuasar/CohereForAI.c4ai-command-r7b-arabic-02-2025-GGUF",
|
67 |
+
filename="CohereForAI.c4ai-command-r7b-arabic-02-2025-Q4_K_M.gguf",
|
68 |
+
local_dir="./models",
|
69 |
+
local_dir_use_symlinks=False
|
70 |
)
|
|
|
71 |
|
72 |
+
# 2) Load the model with llama-cpp via LangChain’s LlamaCpp
|
73 |
+
llm = LlamaCpp(
|
74 |
+
model_path=model_file,
|
75 |
+
flash_attn=False,
|
76 |
+
n_ctx=2048, # or 4096
|
77 |
+
n_batch=512, # or even 256
|
78 |
+
chat_format='chatml'
|
79 |
+
)
|
80 |
|
81 |
+
return llm
|
|
|
82 |
|
|
|
|
|
83 |
|
84 |
+
# --- 4) بناء سلسلة المحادثة ---
|
85 |
+
def build_conversational_chain(vectorstore):
|
86 |
+
"""
|
87 |
+
Creates a ConversationalRetrievalChain using the local llama-cpp-based LLM
|
88 |
+
and a ConversationBufferMemory for multi-turn Q&A.
|
89 |
+
"""
|
90 |
+
llm = load_llm()
|
91 |
|
92 |
+
# We'll store chat history in memory so the chain can handle multi-turn conversations
|
93 |
+
memory = ConversationBufferMemory(
|
94 |
+
memory_key="chat_history",
|
95 |
+
return_messages=True
|
96 |
+
)
|
97 |
|
98 |
+
qa_chain = ConversationalRetrievalChain.from_llm(
|
99 |
+
llm=llm,
|
100 |
+
retriever=vectorstore.as_retriever(search_type="similarity", search_kwargs={"k": 5}),
|
101 |
+
memory=memory,
|
102 |
+
verbose=True
|
103 |
+
)
|
104 |
+
|
105 |
+
return qa_chain
|
106 |
|
|
|
|
|
|
|
|
|
|
|
107 |
|
108 |
+
# --- 5) تنفيذ التطبيق ---
|
109 |
+
vectorstore = build_vectorstore()
|
110 |
+
qa_chain = build_conversational_chain(vectorstore)
|
111 |
+
|
112 |
if "messages" not in st.session_state:
|
113 |
st.session_state["messages"] = [
|
114 |
{"role": "assistant", "content": "👋 مرحبًا! اسألني أي شيء عن إدارة البيانات وحماية البيانات الشخصية!"}
|
|
|
119 |
with st.chat_message(msg["role"]):
|
120 |
st.markdown(f'<div class="rtl">{msg["content"]}</div>', unsafe_allow_html=True)
|
121 |
|
122 |
+
# إدخال المستخدم
|
123 |
user_input = st.chat_input("اكتب سؤالك هنا")
|
124 |
|
|
|
125 |
if user_input:
|
126 |
# عرض رسالة المستخدم
|
127 |
st.session_state["messages"].append({"role": "user", "content": user_input})
|
128 |
with st.chat_message("user"):
|
129 |
st.markdown(f'<div class="rtl">{user_input}</div>', unsafe_allow_html=True)
|
130 |
|
131 |
+
# استدعاء سلسلة المحادثة للحصول على الإجابة
|
132 |
+
response = qa_chain({"question": user_input})
|
|
|
|
|
|
|
|
|
|
|
133 |
|
134 |
# عرض الإجابة
|
135 |
+
answer = response["answer"]
|
136 |
+
st.session_state["messages"].append({"role": "assistant", "content": answer})
|
137 |
with st.chat_message("assistant"):
|
138 |
+
st.markdown(f'<div class="rtl">{answer}</div>', unsafe_allow_html=True)
|