Update app.py

app.py

@@ -2,16 +2,16 @@ import os
 import gc
 import torch
 import gradio as gr
-from langchain.document_loaders import PyMuPDFLoader, TextLoader
+from langchain_community.document_loaders import PyMuPDFLoader, TextLoader
 from langchain.text_splitter import CharacterTextSplitter
-from langchain.vectorstores import FAISS
-from langchain.embeddings import HuggingFaceEmbeddings
+from langchain_community.vectorstores import FAISS
+from langchain_community.embeddings import HuggingFaceEmbeddings
 from langchain.chains import RetrievalQA
-from langchain.llms import HuggingFacePipeline
+from langchain_community.llms import HuggingFacePipeline
 from transformers import pipeline, AutoTokenizer, BitsAndBytesConfig
 from huggingface_hub import login
 
-# Handle HF token securely
+# HF Token handling
 if os.environ.get("HF_TOKEN"):
     login(token=os.environ["HF_TOKEN"])
 
@@ -29,7 +29,7 @@ def load_documents(file_path="study_materials"):
 
 def create_qa_system():
     try:
-        # Load and split documents
+        # Load and process documents
         documents = load_documents()
         if not documents:
             raise ValueError("📚 No study materials found")
@@ -49,13 +49,7 @@ def create_qa_system():
         # Vector store
         db = FAISS.from_documents(texts, embeddings)
 
-        # Quantization config
-        quant_config = BitsAndBytesConfig(
-            load_in_8bit=True,
-            llm_int8_threshold=6.0
-        )
-
-        # LLM setup with optimizations
+        # LLM setup with CPU optimizations
         tokenizer = AutoTokenizer.from_pretrained("google/flan-t5-large")
         pipe = pipeline(
             "text2text-generation",
@@ -68,14 +62,13 @@ def create_qa_system():
             device=-1,  # Force CPU usage
             model_kwargs={
                 "torch_dtype": torch.float16,
-                "quantization_config": quant_config
+                "low_cpu_mem_usage": True
             }
         )
 
         # Memory cleanup
         gc.collect()
 
-        # Create QA system
         return RetrievalQA.from_llm(
             llm=HuggingFacePipeline(pipeline=pipe),
             retriever=db.as_retriever(search_kwargs={"k": 3}),