Update app.py

app.py

@@ -1,20 +1,17 @@
 import streamlit as st
 import os
-from langchain.llms import HuggingFacePipeline
+from huggingface_hub import InferenceClient
 from langchain_community.document_loaders import PDFPlumberLoader
 from langchain.text_splitter import RecursiveCharacterTextSplitter
 from langchain_core.vectorstores import InMemoryVectorStore
 from langchain_core.prompts import ChatPromptTemplate
 from langchain.embeddings import HuggingFaceEmbeddings
-from transformers import pipeline
 
 
 # Set up Hugging Face model and token
-model_name = "mistralai/Mixtral-8x7B-Instruct-v0.1"  # Change to your preferred model
+model = "mistralai/Mixtral-8x7B-Instruct-v0.1"  # You can change to a model of your choice from Hugging Face
 access_token = os.getenv("HF_TOKEN")  # Your Hugging Face API token
-
-# Set up HuggingFace pipeline
-hf_pipeline = pipeline("text-generation", model=model_name, token=access_token)
+client = InferenceClient(model=model, token=access_token)
 
 # Template for response generation
 template = """
@@ -29,7 +26,7 @@ pdfs_directory = '../pdfs'
 os.makedirs(pdfs_directory, exist_ok=True)
 
 # Initialize the embedding model
-embedding = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")  # Choose any model
+embedding = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")  # You can choose any model from Hugging Face
 
 # Initialize the vector store for document indexing
 vector_store = InMemoryVectorStore(embedding=embedding)
@@ -68,11 +65,15 @@ def answer_question(question, documents):
     full_context = f"{context}"
     prompt = ChatPromptTemplate.from_template(template)
     
-    # Use HuggingFacePipeline for generating responses
-    hf_chain = HuggingFacePipeline(pipeline=hf_pipeline)  # Wrap pipeline with HuggingFacePipeline
-    chain = prompt | hf_chain  # Send the prompt to Hugging Face model via HuggingFacePipeline
+    # Format the prompt with the user's question and context
+    question_with_context = prompt.format(question=question, context=full_context)
+    
+    # Use the Hugging Face InferenceClient to generate the response
+    response = client.query(question_with_context)
+    
+    # Assuming the response contains a 'generated_text' field with the model's output
+    return response["generated_text"]
 
-    return chain.invoke({"question": question, "context": full_context})
 
 # Streamlit file uploader for PDF
 uploaded_file = st.file_uploader(