Update app.py

app.py

@@ -1,3 +1,136 @@
+import gradio as gr
+import os
+api_token = os.getenv("HF_TOKEN")
+
+
+from langchain_community.vectorstores import FAISS
+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_community.llms import HuggingFacePipeline
+from langchain.chains import ConversationChain
+from langchain.memory import ConversationBufferMemory
+from langchain_community.llms import HuggingFaceEndpoint
+import torch
+
+list_llm = ["meta-llama/Meta-Llama-3-8B-Instruct", "mistralai/Mistral-7B-Instruct-v0.2"]  
+list_llm_simple = [os.path.basename(llm) for llm in list_llm]
+
+# Load and split PDF document
+def load_doc(list_file_path):
+    # Processing for one document only
+    # loader = PyPDFLoader(file_path)
+    # pages = loader.load()
+    loaders = [PyPDFLoader(x) for x in list_file_path]
+    pages = []
+    for loader in loaders:
+        pages.extend(loader.load())
+    text_splitter = RecursiveCharacterTextSplitter(
+        chunk_size = 1024, 
+        chunk_overlap = 64 
+    )  
+    doc_splits = text_splitter.split_documents(pages)
+    return doc_splits
+
+# Create vector database
+def create_db(splits):
+    embeddings = HuggingFaceEmbeddings()
+    vectordb = FAISS.from_documents(splits, embeddings)
+    return vectordb
+
+
+# Initialize langchain LLM chain
+def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db, progress=gr.Progress()):
+    if llm_model == "meta-llama/Meta-Llama-3-8B-Instruct":
+        llm = HuggingFaceEndpoint(
+            repo_id=llm_model,
+            huggingfacehub_api_token = api_token,
+            temperature = temperature,
+            max_new_tokens = max_tokens,
+            top_k = top_k,
+        )
+    else:
+        llm = HuggingFaceEndpoint(
+            huggingfacehub_api_token = api_token,
+            repo_id=llm_model, 
+            temperature = temperature,
+            max_new_tokens = max_tokens,
+            top_k = top_k,
+        )
+    
+    memory = ConversationBufferMemory(
+        memory_key="chat_history",
+        output_key='answer',
+        return_messages=True
+    )
+
+    retriever=vector_db.as_retriever()
+    qa_chain = ConversationalRetrievalChain.from_llm(
+        llm,
+        retriever=retriever,
+        chain_type="stuff", 
+        memory=memory,
+        return_source_documents=True,
+        verbose=False,
+    )
+    return qa_chain
+
+# Initialize database
+def initialize_database(list_file_obj, progress=gr.Progress()):
+    # Create a list of documents (when valid)
+    list_file_path = [x.name for x in list_file_obj if x is not None]
+    # Load document and create splits
+    doc_splits = load_doc(list_file_path)
+    # Create or load vector database
+    vector_db = create_db(doc_splits)
+    return vector_db, "Database created!"
+
+# Initialize LLM
+def initialize_LLM(llm_option, llm_temperature, max_tokens, top_k, vector_db, progress=gr.Progress()):
+    # print("llm_option",llm_option)
+    llm_name = list_llm[llm_option]
+    print("llm_name: ",llm_name)
+    qa_chain = initialize_llmchain(llm_name, llm_temperature, max_tokens, top_k, vector_db, progress)
+    return qa_chain, "QA chain initialized. Chatbot is ready!"
+
+
+def format_chat_history(message, chat_history):
+    formatted_chat_history = []
+    for user_message, bot_message in chat_history:
+        formatted_chat_history.append(f"User: {user_message}")
+        formatted_chat_history.append(f"Assistant: {bot_message}")
+    return formatted_chat_history
+    
+
+def conversation(qa_chain, message, history):
+    formatted_chat_history = format_chat_history(message, history)
+    # Generate response using QA chain
+    response = qa_chain.invoke({"question": message, "chat_history": formatted_chat_history})
+    response_answer = response["answer"]
+    if response_answer.find("Helpful Answer:") != -1:
+        response_answer = response_answer.split("Helpful Answer:")[-1]
+    response_sources = response["source_documents"]
+    response_source1 = response_sources[0].page_content.strip()
+    response_source2 = response_sources[1].page_content.strip()
+    response_source3 = response_sources[2].page_content.strip()
+    # Langchain sources are zero-based
+    response_source1_page = response_sources[0].metadata["page"] + 1
+    response_source2_page = response_sources[1].metadata["page"] + 1
+    response_source3_page = response_sources[2].metadata["page"] + 1
+    # Append user message and response to chat history
+    new_history = history + [(message, response_answer)]
+    return qa_chain, gr.update(value=""), new_history, response_source1, response_source1_page, response_source2, response_source2_page, response_source3, response_source3_page
+    
+
+def upload_file(file_obj):
+    list_file_path = []
+    for idx, file in enumerate(file_obj):
+        file_path = file_obj.name
+        list_file_path.append(file_path)
+    return list_file_path
+
 def demo():
     custom_css = """
     .horizontal-container {