app.py

import gradio as gr
import os
import tempfile
import shutil
from langchain_community.document_loaders import PyMuPDFLoader, UnstructuredWordDocumentLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores import FAISS
from langchain.chains import RetrievalQA
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
import torch

EMBEDDING_MODEL_NAME = "CAMeL-Lab/bert-base-arabic-camelbert-mix"
QA_MODEL_NAME = "mosaicml/mpt-7b-storywriter"

embedding_model = HuggingFaceEmbeddings(model_name=EMBEDDING_MODEL_NAME)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
qa_tokenizer = AutoTokenizer.from_pretrained(QA_MODEL_NAME, trust_remote_code=True)
qa_model = AutoModelForSeq2SeqLM.from_pretrained(QA_MODEL_NAME, trust_remote_code=True).to(device)

vectordb = None

def load_document(file_path):
    ext = os.path.splitext(file_path)[1].lower()
    if ext == ".pdf":
        loader = PyMuPDFLoader(file_path)
    elif ext in [".doc", ".docx"]:
        loader = UnstructuredWordDocumentLoader(file_path)
    else:
        raise ValueError("صيغة الملف غير مدعومة.")
    return loader.load()

def train_from_documents(documents):
    splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
    texts = splitter.split_documents(documents)
    vectorstore = FAISS.from_documents(texts, embedding_model)
    return vectorstore

def upload_files(files):
    global vectordb
    temp_dir = tempfile.mkdtemp()
    all_docs = []
    
    for file in files:
        file_path = os.path.join(temp_dir, file.name)
        with open(file_path, "wb") as f:
            f.write(file.read())
        docs = load_document(file_path)
        all_docs.extend(docs)
    
    vectordb = train_from_documents(all_docs)
    shutil.rmtree(temp_dir)
    return "✅ النظام جاهز للإجابة على أسئلتك!"

def answer_question(question):
    if vectordb is None:
        return "⚠️ الرجاء رفع الملفات أولاً."
    
    retriever = vectordb.as_retriever(search_kwargs={"k": 5})
    qa_chain = RetrievalQA.from_chain_type(
        llm=None,
        retriever=retriever,
        return_source_documents=True
    )

    relevant_docs = qa_chain.retriever.get_relevant_documents(question)
    context = "\n".join(doc.page_content for doc in relevant_docs)
    
    inputs = qa_tokenizer(
        f"أجب بالعربية فقط بناءً على السياق التالي:\n{context}\nالسؤال: {question}",
        return_tensors="pt",
        truncation=True,
        max_length=1024
    ).to(device)
    
    with torch.no_grad():
        outputs = qa_model.generate(**inputs, max_length=300)
    answer = qa_tokenizer.decode(outputs[0], skip_special_tokens=True)
    return answer

with gr.Blocks(title="محاكاة دماغ المؤلف") as demo:
    with gr.Row():
        with gr.Column():
            gr.Markdown("## 📚 ارفع كتبك هنا")
            file_uploader = gr.File(file_types=[".pdf", ".doc", ".docx"], file_count="multiple")
            upload_button = gr.Button("🚀 ابدأ التدريب")
            training_status = gr.Textbox(label="حالة التدريب", interactive=False)
        
        with gr.Column():
            gr.Markdown("## ❓ اطرح سؤالك")
            question_input = gr.Textbox(label="سؤالك", placeholder="اكتب سؤالك هنا...")
            ask_button = gr.Button("✉️ أرسل السؤال!")
            answer_output = gr.Textbox(label="الإجابة", interactive=False)

    upload_button.click(upload_files, inputs=[file_uploader], outputs=[training_status])
    ask_button.click(answer_question, inputs=[question_input], outputs=[answer_output])

demo.launch(share=True)