Update app.py

app.py

@@ -3,10 +3,9 @@ from langchain_community.document_loaders import PyPDFLoader
 from langchain.text_splitter import RecursiveCharacterTextSplitter
 from langchain_community.vectorstores import Chroma
 from langchain.chains import ConversationalRetrievalChain
-from langchain_community.embeddings import HuggingFaceEmbeddings
+from langchain_huggingface import HuggingFaceEmbeddings
 from langchain.chains import ConversationChain
 from langchain.memory import ConversationBufferMemory
-from langchain_community.llms import HuggingFaceEndpoint
 
 from pathlib import Path
 import chromadb
@@ -15,7 +14,7 @@ from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
 import re
 
 # Constants
-LLM_MODEL = "sentence-transformers/all-MiniLM-L6-v2"
+LLM_MODEL = "t5-small"  # Changed to a Seq2Seq model compatible with AutoModelForSeq2SeqLM
 DB_CHUNK_SIZE = 512
 CHUNK_OVERLAP = 24
 TEMPERATURE = 0.1
@@ -33,7 +32,7 @@ def load_doc(pdf_url, chunk_size, chunk_overlap):
 
 # Create vector database
 def create_db(splits, collection_name):
-    embedding = HuggingFaceEmbeddings(model_name=LLM_MODEL)
+    embedding = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
     new_client = chromadb.EphemeralClient()
     vectordb = Chroma.from_documents(
         documents=splits,
@@ -61,7 +60,7 @@ def initialize_llmchain(llm_model, vector_db, progress=gr.Progress()):
     retriever = vector_db.as_retriever()
     progress(0.8, desc="Defining retrieval chain...")
     qa_chain = ConversationalRetrievalChain.from_llm(
-        pipe,
+        llm=pipe,
         retriever=retriever,
         chain_type="stuff",
         memory=memory,
@@ -111,7 +110,7 @@ def conversation(qa_chain, message, history):
     formatted_chat_history = format_chat_history(message, history)
     response = qa_chain({"question": message, "chat_history": formatted_chat_history})
     response_answer = response["answer"]
-    if response_answer.find("Helpful Answer:") != -1:
+    if "Helpful Answer:" in response_answer:
         response_answer = response_answer.split("Helpful Answer:")[-1]
     response_sources = response["source_documents"]
     response_source1 = response_sources[0].page_content.strip()
@@ -135,7 +134,7 @@ def demo():
             <h3>Ask any questions about your PDF documents</h3>""")
         gr.Markdown(
             """<b>Note:</b> This AI assistant, using Langchain and open-source LLMs, performs retrieval-augmented generation (RAG) from your PDF documents. \
-            The user interface explicitely shows multiple steps to help understand the RAG workflow. 
+            The user interface explicitly shows multiple steps to help understand the RAG workflow. 
             This chatbot takes past questions into account when generating answers (via conversational memory), and includes document references for clarity purposes.<br>
             <br><b>Warning:</b> This space uses the free CPU Basic hardware from Hugging Face. Some steps and LLM models used below (free inference endpoints) can take some time to generate a reply.
             """)