Upload 3 files

app.py

@@ -0,0 +1,19 @@
+import gradio as gr
+from rag_pipeline import load_rag_chain
+
+rag_chain = load_rag_chain()
+
+def ask_question(query):
+    result = rag_chain.invoke(query)
+    return result['result']
+
+iface = gr.Interface(
+    fn=ask_question,
+    inputs=gr.Textbox(lines=3, placeholder="اكتب سؤالك هنا...", label="سؤالك", rtl=True),
+    outputs=gr.Textbox(label="الإجابة", rtl=True),
+    title="🧠 روبوت دردشة عربي باستخدام ملفات PDF",
+    description="اكتب سؤالك بالعربية (يدعم اللهجة المصرية)، وسنبحث عن الإجابة داخل الملفات التي قمت بتحميلها."
+)
+
+if __name__ == "__main__":
+    iface.launch()

rag_pipeline.py

@@ -0,0 +1,44 @@
+from pathlib import Path
+from langchain.chains import RetrievalQA
+from transformers import pipeline, AutoTokenizer
+from langchain_community.vectorstores import Chroma
+from langchain_community.document_loaders import PyMuPDFLoader
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain_huggingface import HuggingFaceEmbeddings, HuggingFacePipeline
+
+def load_documents(pdf_dir):
+    docs = []
+    for pdf_file in Path(pdf_dir).glob("*.pdf"):
+        loader = PyMuPDFLoader(str(pdf_file))
+        docs.extend(loader.load())
+    return docs
+
+def load_rag_chain():
+    pdf_dir = Path("data")
+    pdf_dir.mkdir(parents=True, exist_ok=True)
+
+    raw_docs = load_documents(pdf_dir)
+    splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
+    pages = splitter.split_documents(raw_docs)
+
+    embeddings = HuggingFaceEmbeddings(
+        model_name="sentence-transformers/LaBSE",
+        model_kwargs={"device": "cpu"},
+    )
+
+    vectordb_dir = "chroma_db"
+    vectordb = Chroma.from_documents(pages, embeddings, persist_directory=vectordb_dir)
+    retriever = vectordb.as_retriever(search_type="mmr", search_kwargs={"k": 5})
+
+    hf_pipeline = pipeline(
+        "text2text-generation",
+        model="csebuetnlp/mT5_small_finetuned_squad",
+        tokenizer=AutoTokenizer.from_pretrained("csebuetnlp/mT5_small_finetuned_squad"),
+        max_new_tokens=512,
+        temperature=0.3,
+        device=-1
+    )
+    llm = HuggingFacePipeline(pipeline=hf_pipeline)
+
+    qa_chain = RetrievalQA.from_llm(llm=llm, retriever=retriever)
+    return qa_chain

requirements.txt

@@ -0,0 +1,8 @@
+gradio
+transformers
+sentence-transformers
+langchain
+langchain-community
+langchain-huggingface
+chromadb
+pymupdf