Upload 6 files

app.py

@@ -0,0 +1,111 @@
+import streamlit as st
+import os
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
+from langchain_community.document_loaders import PyPDFLoader
+from langchain_experimental.text_splitter import SemanticChunker
+from langchain_huggingface import HuggingFaceEmbeddings
+from langchain_community.vectorstores import FAISS
+from langchain.memory import ConversationBufferMemory
+
+# --- 1) إعداد الصفحة ---
+st.title("💬 المحادثة التفاعلية - إدارة البيانات وحماية البيانات الشخصية")
+local_file = "Policies001.pdf"
+
+index_folder = "faiss_index"
+
+# إضافة CSS مخصص لدعم النصوص من اليمين لليسار
+st.markdown(
+    """
+    <style>
+        .rtl {
+            direction: rtl;
+            text-align: right;
+        }
+    </style>
+    """,
+    unsafe_allow_html=True
+)
+
+# --- 2) تحميل أو بناء قاعدة بيانات FAISS ---
+embeddings = HuggingFaceEmbeddings(
+    model_name="CAMeL-Lab/bert-base-arabic-camelbert-mix",
+    model_kwargs={"trust_remote_code": True}
+)
+
+if os.path.exists(index_folder):
+    # تحميل قاعدة البيانات إذا كانت موجودة
+    vectorstore = FAISS.load_local(index_folder, embeddings, allow_dangerous_deserialization=True)
+else:
+    # تحميل PDF وتقسيم النصوص
+    loader = PyPDFLoader(local_file)
+    documents = loader.load()
+
+    text_splitter = SemanticChunker(
+        embeddings=embeddings,
+        breakpoint_threshold_type='percentile',
+        breakpoint_threshold_amount=90
+    )
+    chunked_docs = text_splitter.split_documents(documents)
+
+    # إنشاء قاعدة بيانات FAISS
+    vectorstore = FAISS.from_documents(chunked_docs, embeddings)
+    vectorstore.save_local(index_folder)
+
+# --- 3) إعداد المسترجع ---
+retriever = vectorstore.as_retriever(search_type="similarity", search_kwargs={"k": 5})
+
+# --- 4) إعداد نموذج النص ---
+model_name = "CohereForAI/c4ai-command-r7b-arabic-02-2025"  # اسم النموذج
+
+# التأكد من وجود توكن Hugging Face
+hf_token = os.getenv("HF_TOKEN")
+if hf_token is None:
+    st.error("Hugging Face token not found. Please set the 'HF_TOKEN' environment variable.")
+    st.stop()
+
+# تحميل النموذج والمحول
+tokenizer = AutoTokenizer.from_pretrained(model_name, use_auth_token=hf_token)
+model = AutoModelForCausalLM.from_pretrained(model_name, use_auth_token=hf_token)
+
+# إعداد pipeline لتوليد النصوص
+qa_pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer, device=0)
+
+# --- 5) إعداد الذاكرة ---
+memory = ConversationBufferMemory(
+    memory_key="chat_history",
+    return_messages=True
+)
+
+# --- 6) إدارة رسائل المستخدم ---
+if "messages" not in st.session_state:
+    st.session_state["messages"] = [
+        {"role": "assistant", "content": "👋 مرحبًا! اسألني أي شيء عن إدارة البيانات وحماية البيانات الشخصية!"}
+    ]
+
+# عرض الرسائل الحالية
+for msg in st.session_state["messages"]:
+    with st.chat_message(msg["role"]):
+        st.markdown(f'<div class="rtl">{msg["content"]}</div>', unsafe_allow_html=True)
+
+# --- 7) إدخال المستخدم ---
+user_input = st.chat_input("اكتب سؤالك هنا")
+
+# --- 8) معالجة رسالة المستخدم ---
+if user_input:
+    # عرض رسالة المستخدم
+    st.session_state["messages"].append({"role": "user", "content": user_input})
+    with st.chat_message("user"):
+        st.markdown(f'<div class="rtl">{user_input}</div>', unsafe_allow_html=True)
+
+    # استرجاع المستندات ذات الصلة
+    retrieved_docs = retriever.get_relevant_documents(user_input)
+    context = "\n".join([doc.page_content for doc in retrieved_docs])
+    full_input = f"السياق:\n{context}\n\nالسؤال:\n{user_input}"
+
+    # توليد الإجابة باستخدام النموذج
+    response = qa_pipeline(full_input, max_length=500, num_return_sequences=1)[0]["generated_text"]
+
+    # عرض الإجابة
+    st.session_state["messages"].append({"role": "assistant", "content": response})
+    with st.chat_message("assistant"):
+        st.markdown(f'<div class="rtl">{response}</div>', unsafe_allow_html=True)

chain_setup.py

@@ -0,0 +1,50 @@
+import os
+from huggingface_hub import hf_hub_download
+from langchain.llms import LlamaCpp
+from langchain.chains import ConversationalRetrievalChain
+from langchain.memory import ConversationBufferMemory
+
+def load_llm():
+    """
+    Downloads the Q4_K_M GGUF model from mobeidat's Hugging Face repository and loads it via llama-cpp.
+    """
+    # 1) Download the GGUF model from Hugging Face
+    model_file = hf_hub_download(
+        repo_id="bartowski/c4ai-command-r7b-12-2024-GGUF",
+        filename="c4ai-command-r7b-12-2024-Q4_K_M.gguf",
+        local_dir="./models",
+        local_dir_use_symlinks=False
+    )
+
+    # 2) Load the model with llama-cpp via LangChain’s LlamaCpp
+    llm = LlamaCpp(
+        model_path=model_file,
+        flash_attn=False,
+        n_ctx=2048,  # or 4096 depending on your needs
+        n_batch=512,  # or even 256 depending on your hardware
+        chat_format='chatml'
+    )
+
+    return llm
+
+def build_conversational_chain(vectorstore):
+    """
+    Creates a ConversationalRetrievalChain using the local llama-cpp-based LLM 
+    and a ConversationBufferMemory for multi-turn Q&A.
+    """
+    llm = load_llm()
+
+    # We'll store chat history in memory so the chain can handle multi-turn conversations
+    memory = ConversationBufferMemory(
+        memory_key="chat_history",
+        return_messages=True
+    )
+
+    qa_chain = ConversationalRetrievalChain.from_llm(
+        llm=llm,
+        retriever=vectorstore.as_retriever(search_type="similarity", search_kwargs={"k": 5}),
+        memory=memory,
+        verbose=True
+    )
+
+    return qa_chain

embedding.py

@@ -0,0 +1,13 @@
+# embedding.py
+
+from langchain_huggingface import HuggingFaceEmbeddings
+
+def load_embeddings():
+    """
+    Returns a HuggingFaceEmbeddings instance.
+    """
+    embeddings = HuggingFaceEmbeddings(
+        model_name="CAMeL-Lab/bert-base-arabic-camelbert-mix",
+        model_kwargs={"trust_remote_code": True}
+    )
+    return embeddings

runtime.txt

@@ -0,0 +1 @@
+python-3.12

streamlit_app.py

@@ -0,0 +1,58 @@
+# app.py
+
+import streamlit as st
+import os
+
+# Local imports
+from embedding import load_embeddings
+from vectorstore import load_or_build_vectorstore
+from chain_setup import build_conversational_chain
+
+def main():
+    st.title("💬 المحادثة التفاعلية - ادارة البيانات و حماية البيانات الشخصية")
+
+    # Paths and constants
+    local_file = "Policies001.pdf"
+    index_folder = "faiss_index"
+
+    # Step 1: Load Embeddings
+    embeddings = load_embeddings()
+
+    # Step 2: Build or load VectorStore
+    vectorstore = load_or_build_vectorstore(local_file, index_folder, embeddings)
+
+    # Step 3: Build the Conversational Retrieval Chain
+    qa_chain = build_conversational_chain(vectorstore)
+
+    # Step 4: Session State for UI Chat
+    if "messages" not in st.session_state:
+        st.session_state["messages"] = [
+            {"role": "assistant", "content": "👋 مرحبًا! اسألني أي شيء عن إدارة البيانات وحماية البيانات الشخصية"}
+        ]
+
+    # Display existing messages
+    for msg in st.session_state["messages"]:
+        with st.chat_message(msg["role"]):
+            st.markdown(msg["content"])
+
+    # Step 5: Chat Input
+    user_input = st.chat_input("Type your question...")
+
+    # Step 6: Process user input
+    if user_input:
+        # a) Display user message
+        st.session_state["messages"].append({"role": "user", "content": user_input})
+        with st.chat_message("user"):
+            st.markdown(user_input)
+
+        # b) Run chain
+        response_dict = qa_chain({"question": user_input})
+        answer = response_dict["answer"]
+
+        # c) Display assistant response
+        st.session_state["messages"].append({"role": "assistant", "content": answer})
+        with st.chat_message("assistant"):
+            st.markdown(answer)
+
+if __name__ == "__main__":
+    main()

vectorstore.py

@@ -0,0 +1,32 @@
+# vectorstore.py
+
+import os
+from langchain_community.document_loaders import PyPDFLoader
+from langchain_experimental.text_splitter import SemanticChunker
+from langchain_community.vectorstores import FAISS
+
+def load_or_build_vectorstore(local_file: str, index_folder: str, embeddings):
+    """
+    Loads a local FAISS index if it exists; otherwise,
+    builds a new index from the specified PDF file.
+    """
+    if os.path.exists(index_folder):
+        print("Loading existing FAISS index from disk...")
+        vectorstore = FAISS.load_local(index_folder, embeddings, allow_dangerous_deserialization=True)
+    else:
+        print("Building a new FAISS index...")
+        loader = PyPDFLoader(local_file)
+        documents = loader.load()
+
+        text_splitter = SemanticChunker(
+            embeddings=embeddings,
+            breakpoint_threshold_type='percentile',
+            breakpoint_threshold_amount=90
+        )
+        chunked_docs = text_splitter.split_documents(documents)
+        print(f"Document split into {len(chunked_docs)} chunks.")
+
+        vectorstore = FAISS.from_documents(chunked_docs, embeddings)
+        vectorstore.save_local(index_folder)
+
+    return vectorstore