Update app.py

app.py

@@ -1,6 +1,6 @@
 import streamlit as st
 import os
-from transformers import pipeline
+from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM
 
 # --- LANGCHAIN IMPORTS ---
 from langchain_community.document_loaders import PyPDFLoader
@@ -51,14 +51,14 @@ else:
 # 3) CREATE RETRIEVER
 retriever = vectorstore.as_retriever(search_type="similarity", search_kwargs={"k": 5})
 
-# 4) SET UP HF TRANSFORMERS PIPELINE
-# Use Hugging Face's transformers pipeline for text generation
-hf_pipeline = pipeline(
-    "text-generation",
-    model="UBC-NLP/AraT5-base",  # Arabic-friendly model
-    tokenizer="UBC-NLP/AraT5-base",
-    device=-1
-)
+# 4) LOAD HF TRANSFORMERS MODEL AND PIPELINE
+# Use the AutoModelForSeq2SeqLM for T5-based models
+model_name = "UBC-NLP/AraT5-base"  # Replace with any model suitable for Arabic tasks
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
+
+# Define a text2text-generation pipeline for T5
+hf_pipeline = pipeline("text2text-generation", model=model, tokenizer=tokenizer, device=-1)
 
 # Define a function to generate responses
 def generate_response(question, retriever):
@@ -70,7 +70,7 @@ def generate_response(question, retriever):
     context = " ".join([doc.page_content for doc in retrieved_docs])
 
     # Generate the response using the HF pipeline
-    input_text = f"{question} Context: {context}"
+    input_text = f"سؤال: {question}\nالنص: {context}\nالإجابة:"
     response = hf_pipeline(input_text, max_length=200, num_return_sequences=1)
     return response[0]["generated_text"]