Upload 7 files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+faiss_index/index.faiss filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,67 @@
+import streamlit as st
+import os
+
+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 - إدارة البيانات وحماية البيانات الشخصية")
+
+    # Apply RTL custom CSS for right-to-left text alignment
+    st.markdown(
+        """
+        <style>
+            .rtl {
+                direction: rtl;
+                text-align: right;
+            }
+        </style>
+        """,
+        unsafe_allow_html=True
+    )
+
+    # Paths and constants
+    local_file = "Policies001.pdf"
+    index_folder = "faiss_index"
+
+    # Step 1: Load Arabic Embeddings
+    embeddings = load_embeddings()
+
+    # Step 2: Build or load the 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 with RTL styling
+    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)
+
+    # Step 5: Chat Input
+    user_input = st.chat_input("اكتب سؤالك هنا")
+
+    if user_input:
+        # a) Display user message
+        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)
+
+        # b) Run chain
+        response = qa_chain({"question": user_input})
+        answer = response["answer"]
+
+        # c) Display assistant response
+        st.session_state["messages"].append({"role": "assistant", "content": answer})
+        with st.chat_message("assistant"):
+            st.markdown(f'<div class="rtl">{answer}</div>', unsafe_allow_html=True)
+
+if __name__ == "__main__":
+    main()

chain_setup.py

@@ -0,0 +1,45 @@
+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 GGUF model for Arabic and loads it via llama-cpp.
+    """
+    model_file = hf_hub_download(
+        repo_id="Moe98/cohere-r7b-arabic-02-2025-GGUF",
+        filename="cohere-r7b-02-2025-Q8_0.gguf",
+        local_dir="./models",
+        local_dir_use_symlinks=False
+    )
+
+    llm = LlamaCpp(
+        model_path=model_file,
+        flash_attn=False,
+        n_ctx=2048,
+        n_batch=512,
+        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()
+
+    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 +1,11 @@
+from langchain_huggingface import HuggingFaceEmbeddings
+
+def load_embeddings():
+    """
+    Returns a HuggingFaceEmbeddings instance for Arabic.
+    """
+    embeddings = HuggingFaceEmbeddings(
+        model_name="CAMeL-Lab/bert-base-arabic-camelbert-mix",
+        model_kwargs={"trust_remote_code": True}
+    )
+    return embeddings

faiss_index/index.faiss

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:126eb6b6953e61d0a121fb87a2856d2a716f75c5f7f6a60e3d6c05bb95d0c836
+size 1520685

faiss_index/index.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a1548b887e6e79289fe03498c29322fac5030208daf71270666cca2e0293d19
+size 277539

requirements.txt

@@ -0,0 +1,16 @@
+streamlit==1.25.0
+langchain
+langchain_community
+torch
+transformers
+sentence-transformers
+accelerate
+pypdf
+ollama
+langchain_experimental
+faiss-cpu
+langchain_huggingface
+einops
+huggingface-hub
+sentencepiece
+llama-cpp-python

vectorstore.py

@@ -0,0 +1,30 @@
+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