Update app.py

app.py

@@ -4,23 +4,17 @@ from langchain_community.document_loaders import PyPDFLoader
 from langchain_text_splitters import RecursiveCharacterTextSplitter
 from langchain_community.embeddings.fastembed import FastEmbedEmbeddings
 from langchain_community.vectorstores import FAISS
-from langchain_huggingface import HuggingFaceEndpoint
+from huggingface_hub import InferenceClient
 from langchain_core.prompts import ChatPromptTemplate
 
-# --- 1. Model Setup using HF Inference API ---
-# Get the HF token from environment variables (set in Space secrets)
+# --- 1. Model Setup using HF Inference Client ---
 HF_TOKEN = os.environ.get("HF_TOKEN", "")
 
 if not HF_TOKEN:
     print("⚠️ Warning: HF_TOKEN not set. The app may not work properly.")
-    print("Please add your Hugging Face token in Space Settings > Repository secrets")
 
-llm = HuggingFaceEndpoint(
-    repo_id="meta-llama/Meta-Llama-3-8B-Instruct",
-    temperature=0.7,
-    max_new_tokens=2000,
-    huggingfacehub_api_token=HF_TOKEN
-)
+# Use InferenceClient directly instead of LangChain wrapper
+client = InferenceClient(token=HF_TOKEN)
 
 # --- 2. The Core Logic ---
 def generate_question_paper(pdf_file, difficulty, num_questions):
@@ -54,11 +48,11 @@ def generate_question_paper(pdf_file, difficulty, num_questions):
         context_docs = retriever.invoke("Key concepts and definitions")
         context_text = "\n\n".join([doc.page_content for doc in context_docs])
         
-        # E. Prompt
-        template = """You are an expert academic examiner. Create a formal Question Paper based ONLY on the context provided below.
+        # E. Create Prompt
+        prompt = f"""You are an expert academic examiner. Create a formal Question Paper based ONLY on the context provided below.
 
 CONTEXT:
-{context}
+{context_text}
 
 INSTRUCTIONS:
 - Difficulty: {difficulty}
@@ -71,25 +65,23 @@ INSTRUCTIONS:
 
 Do not output conversational text. Output ONLY the exam paper in a well-formatted structure."""
         
-        prompt = ChatPromptTemplate.from_template(template)
-        
-        # F. Generate
-        chain = prompt | llm
-        response = chain.invoke({
-            "context": context_text, 
-            "difficulty": difficulty,
-            "num_questions": num_questions
-        })
+        # F. Generate using Mistral (no license needed and works better)
+        response = client.text_generation(
+            prompt,
+            model="mistralai/Mistral-7B-Instruct-v0.3",
+            max_new_tokens=2000,
+            temperature=0.7,
+        )
         
         return response
 
     except Exception as e:
-        return f"❌ Error processing PDF: {str(e)}\n\nPlease check:\n1. PDF is valid and contains text\n2. HF_TOKEN is correctly set\n3. You have access to Llama 3"
+        return f"❌ Error: {str(e)}\n\nPlease check:\n1. PDF is valid and contains text\n2. HF_TOKEN is correctly set in Space secrets\n3. Try again or contact support"
 
 # --- 3. The UI ---
-with gr.Blocks(title="AI Question Paper Generator") as demo:
+with gr.Blocks(title="AI Question Paper Generator", theme=gr.themes.Soft(primary_hue="blue")) as demo:
     gr.Markdown("# 📄 AI Question Paper Generator")
-    gr.Markdown("Powered by **Llama 3 (8B)** via Hugging Face Inference API")
+    gr.Markdown("Powered by **Mistral 7B** via Hugging Face Inference API")
     gr.Markdown("⚡ Fast • 🎯 Accurate • 📚 Context-Aware")
     
     with gr.Row():
@@ -131,8 +123,8 @@ with gr.Blocks(title="AI Question Paper Generator") as demo:
     
     gr.Markdown("""
     ---
-    **Note:** This app requires a Hugging Face token with access to Llama 3. 
-    Set `HF_TOKEN` in your Space's repository secrets.
+    **Note:** Set `HF_TOKEN` in your Space's Settings → Repository secrets.
+    Get your token from https://huggingface.co/settings/tokens
     """)
 
 if __name__ == "__main__":