Update app.py

app.py

@@ -1,35 +1,284 @@
-from pathlib import Path
-from typing import Union
-
-from pypdf import PdfReader
-from transformers import pipeline
 import gradio as gr
-from langchain.llms import HuggingFaceHub
 import os
-from getpass import getpass
+
+from langchain.document_loaders import PyPDFLoader
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.vectorstores import Chroma
+from langchain.chains import ChatVectorDBChain
+from langchain.embeddings import OpenAIEmbeddings
+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
+from langchain_community.chat_models.openai import ChatOpenAI
+from langchain.chains import ConversationalRetrievalChain
+from langchain.llms import OpenAI
+
+from pathlib import Path
+import chromadb
+from unidecode import unidecode
+
+from transformers import AutoTokenizer
+import transformers
+import torch
+import tqdm 
+import accelerate
+import re
+
+
 TOKEN = os.getenv('HUGGING_FACE_HUB_TOKEN')
 
 
-question_answerer =pipeline("text-generation", model="HuggingFaceH4/zephyr-7b-alpha")
+    
+# Load PDF document and create doc splits
+def load_doc(list_file_path, chunk_size, chunk_overlap):
+    # 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 = 600, chunk_overlap = 50)
+    text_splitter = RecursiveCharacterTextSplitter(
+        chunk_size = chunk_size, 
+        chunk_overlap = chunk_overlap)
+    doc_splits = text_splitter.split_documents(pages)
+    return doc_splits
+
+
+# Initialize your vector database with OpenAIEmbeddings and persist it
+def create_db(documents, collection_name, persist_directory="."):
+    embeddings = OpenAIEmbeddings()
+    vectordb = Chroma.from_documents(
+        documents=documents,
+        embedding=embeddings,
+        collection_name=collection_name,
+        persist_directory=persist_directory
+    )
+    vectordb.persist()
+    return vectordb
+
+# Load vector database
+def load_db():
+    embedding = HuggingFaceEmbeddings()
+    vectordb = Chroma(
+        # persist_directory=default_persist_directory, 
+        embedding_function=embedding)
+    return vectordb
+
+
+
+
+# Assuming vectordb is correctly initialized and persisted as shown above
+def initialize_llmchain(vectordb, model_name="gpt-3.5-turbo", temperature=0.9):
+    chat_model = ChatOpenAI(temperature=temperature, model_name=model_name)
+    chat_vector_db_chain = ChatVectorDBChain.from_llm(
+        llm=chat_model,
+        vectordb=vectordb,
+        return_source_documents=True
+    )
+    return chat_vector_db_chain
+
+# Example usage
+if __name__ == "__main__":
+    # Suppose `documents` is a list of your documents and `collection_name` is your desired collection name
+    documents = [...]  # your documents here
+    collection_name = "your_collection_name"
+    vectordb = create_db(documents, collection_name)
+
+    # Initialize the ChatVectorDBChain with the vector database
+    chat_vector_db_chain = initialize_chat_vector_db_chain(vectordb)
+    
+    # Use the chain to process a query
+    query = "your query here"
+    result = chat_vector_db_chain({"question": query, "chat_history": ""})
+    
+    print("Answer:")
+    print(result["answer"])
+# Generate collection name for vector database
+#  - Use filepath as input, ensuring unicode text
+def create_collection_name(filepath):
+    # Extract filename without extension
+    collection_name = Path(filepath).stem
+    # Fix potential issues from naming convention
+    ## Remove space
+    collection_name = collection_name.replace(" ","-") 
+    ## ASCII transliterations of Unicode text
+    collection_name = unidecode(collection_name)
+    ## Remove special characters
+    #collection_name = re.findall("[\dA-Za-z]*", collection_name)[0]
+    collection_name = re.sub('[^A-Za-z0-9]+', '-', collection_name)
+    ## Limit length to 50 characters
+    collection_name = collection_name[:50]
+    ## Minimum length of 3 characters
+    if len(collection_name) < 3:
+        collection_name = collection_name + 'xyz'
+    print('Filepath: ', filepath)
+    print('Collection name: ', collection_name)
+    return collection_name
+
+
+# Initialize database
+def initialize_database(list_file_obj, chunk_size, chunk_overlap, progress=gr.Progress()):
+    # Create list of documents (when valid)
+    list_file_path = [x.name for x in list_file_obj if x is not None]
+    # Create collection_name for vector database
+    progress(0.1, desc="Creating collection name...")
+    collection_name = create_collection_name(list_file_path[0])
+    progress(0.25, desc="Loading document...")
+    # Load document and create splits
+    doc_splits = load_doc(list_file_path, chunk_size, chunk_overlap)
+    # Create or load vector database
+    progress(0.5, desc="Generating vector database...")
+    # global vector_db
+    vector_db = create_db(doc_splits, collection_name)
+    progress(0.9, desc="Done!")
+    return vector_db, collection_name, "Complete!"
+
+
+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, "Complete!"
+
+
+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)
+    #print("formatted_chat_history",formatted_chat_history)
+   
+    # Generate response using QA chain
+    response = qa_chain({"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
+    # print ('chat response: ', response_answer)
+    # print('DB source', response_sources)
+    
+    # Append user message and response to chat history
+    new_history = history + [(message, response_answer)]
+    # return gr.update(value=""), new_history, response_sources[0], response_sources[1] 
+    return qa_chain, gr.update(value=""), new_history, response_source1, response_source1_page, response_source2, response_source2_page, response_source3, response_source3_page
+    
 
-def get_text_from_pdf(pdf_file: Union[str, Path]) -> str:
-    """Read the PDF from the given path and return a string with its entire content."""
-    reader = PdfReader(pdf_file)
+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)
+    # print(file_path)
+    # initialize_database(file_path, progress)
+    return list_file_path
 
-    # Extract text from all pages
-    full_text = ""
-    for page in reader.pages:
-        full_text += page.extract_text()
-    return full_text
+        
+def demo():
+    with gr.Blocks(theme="base") as demo:
+        vector_db = gr.State()
+        qa_chain = gr.State()
+        collection_name = gr.State()
+        
+        gr.Markdown(
+        """PDF-based chatbot (by Dr. Aloke Upadhaya)</center></h2>
+        <h3>Ask any questions about your PDF documents, along with follow-ups</h3>
+        """)
+            
+        with gr.Tab("Step 1 - Document pre-processing"):
+            with gr.Row():
+                document = gr.Files(height=100, file_count="multiple", file_types=["pdf"], interactive=True, label="Upload your PDF documents (single or multiple)")
+                # upload_btn = gr.UploadButton("Loading document...", height=100, file_count="multiple", file_types=["pdf"], scale=1)
+            with gr.Row():
+                db_btn = gr.Radio(["ChromaDB"], label="Vector database type", value = "ChromaDB", type="index", info="Choose your vector database")
+            with gr.Accordion("Advanced options - Document text splitter", open=False):
+                with gr.Row():
+                    slider_chunk_size = gr.Slider(minimum = 100, maximum = 1000, value=600, step=20, label="Chunk size", info="Chunk size", interactive=True)
+                with gr.Row():
+                    slider_chunk_overlap = gr.Slider(minimum = 10, maximum = 200, value=40, step=10, label="Chunk overlap", info="Chunk overlap", interactive=True)
+            with gr.Row():
+                db_progress = gr.Textbox(label="Vector database initialization", value="None")
+            with gr.Row():
+                db_btn = gr.Button("Generate vector database...")
+            
+        with gr.Tab("Step 2 - QA chain initialization"):
+            with gr.Row():
+                llm_btn = gr.Radio(list_llm_simple, \
+                    label="LLM models", value = list_llm_simple[0], type="index", info="Choose your LLM model")
+            with gr.Accordion("Advanced options - LLM model", open=False):
+                with gr.Row():
+                    slider_temperature = gr.Slider(minimum = 0.0, maximum = 1.0, value=0.7, step=0.1, label="Temperature", info="Model temperature", interactive=True)
+                with gr.Row():
+                    slider_maxtokens = gr.Slider(minimum = 224, maximum = 4096, value=1024, step=32, label="Max Tokens", info="Model max tokens", interactive=True)
+                with gr.Row():
+                    slider_topk = gr.Slider(minimum = 1, maximum = 10, value=3, step=1, label="top-k samples", info="Model top-k samples", interactive=True)
+            with gr.Row():
+                llm_progress = gr.Textbox(value="None",label="QA chain initialization")
+            with gr.Row():
+                qachain_btn = gr.Button("Initialize question-answering chain...")
 
+        with gr.Tab("Step 3 - Conversation with chatbot"):
+            chatbot = gr.Chatbot(height=300)
+            with gr.Accordion("Advanced - Document references", open=False):
+                with gr.Row():
+                    doc_source1 = gr.Textbox(label="Reference 1", lines=2, container=True, scale=20)
+                    source1_page = gr.Number(label="Page", scale=1)
+                with gr.Row():
+                    doc_source2 = gr.Textbox(label="Reference 2", lines=2, container=True, scale=20)
+                    source2_page = gr.Number(label="Page", scale=1)
+                with gr.Row():
+                    doc_source3 = gr.Textbox(label="Reference 3", lines=2, container=True, scale=20)
+                    source3_page = gr.Number(label="Page", scale=1)
+            with gr.Row():
+                msg = gr.Textbox(placeholder="Type message", container=True)
+            with gr.Row():
+                submit_btn = gr.Button("Submit")
+                clear_btn = gr.ClearButton([msg, chatbot])
+            
+        # Preprocessing events
+        #upload_btn.upload(upload_file, inputs=[upload_btn], outputs=[document])
+        db_btn.click(initialize_database, \
+            inputs=[document, slider_chunk_size, slider_chunk_overlap], \
+            outputs=[vector_db, collection_name, db_progress])
+        qachain_btn.click(initialize_LLM, \
+            inputs=[llm_btn, slider_temperature, slider_maxtokens, slider_topk, vector_db], \
+            outputs=[qa_chain, llm_progress]).then(lambda:[None,"",0,"",0,"",0], \
+            inputs=None, \
+            outputs=[chatbot, doc_source1, source1_page, doc_source2, source2_page, doc_source3, source3_page], \
+            queue=False)
 
-def answer_doc_question(pdf_file, question):
-    pdf_text = get_text_from_pdf(pdf_file)
-    answer = question_answerer(question, pdf_text)
-    return answer["answer"]
+        # Chatbot events
+        msg.submit(conversation, \
+            inputs=[qa_chain, msg, chatbot], \
+            outputs=[qa_chain, msg, chatbot, doc_source1, source1_page, doc_source2, source2_page, doc_source3, source3_page], \
+            queue=False)
+        submit_btn.click(conversation, \
+            inputs=[qa_chain, msg, chatbot], \
+            outputs=[qa_chain, msg, chatbot, doc_source1, source1_page, doc_source2, source2_page, doc_source3, source3_page], \
+            queue=False)
+        clear_btn.click(lambda:[None,"",0,"",0,"",0], \
+            inputs=None, \
+            outputs=[chatbot, doc_source1, source1_page, doc_source2, source2_page, doc_source3, source3_page], \
+            queue=False)
+    demo.queue().launch(debug=True)
 
 
-pdf_input = gr.File(file_types=[".pdf"], label="Upload a PDF document and ask a question about it.")
-question = gr.Textbox(label="Type a question regarding the uploaded document here.")
-gr.Interface(fn=answer_doc_question, inputs=[pdf_input, question], outputs="text").launch()
+if __name__ == "__main__":
+    demo()