Update app.py

app.py

@@ -1,151 +1,238 @@
-from typing import Any
 import gradio as gr
-from langchain.embeddings.openai import OpenAIEmbeddings
-from langchain.vectorstores import Chroma
-
+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.chat_models import ChatOpenAI
+from langchain_community.embeddings import OpenAIEmbeddings
+from langchain_community.llms import ChatOpenAI
+from langchain.memory import ConversationBufferMemory
+from pathlib import Path
+import chromadb
+from unidecode import unidecode
+import re
 
-from langchain.document_loaders import PyPDFLoader
+# Assume list_llm and other previously defined variables and functions are available
 
-import fitz
-from PIL import Image
+def get_openai_api_key():
+    """Function to prompt the user to input their OpenAI API key."""
+    api_key = input("Please enter your OpenAI API key: ")
+    return api_key
+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
+def create_db(splits, collection_name, api_key):
+    """Adjusted to include OpenAI API key for embeddings."""
+    embedding = OpenAIEmbeddings(api_key=api_key)  # Utilize the OpenAI API key
+    new_client = chromadb.EphemeralClient()
+    vectordb = Chroma.from_documents(
+        documents=splits,
+        embedding=embedding,
+        client=new_client,
+        collection_name=collection_name,
+    )
+    return vectordb
+def initialize_llmchain(llm_model, temperature, max_tokens, top_k, vector_db, api_key):
+    """Adjusted to include OpenAI API key for the LLM initialization."""
+    llm = ChatOpenAI(api_key=api_key, temperature=temperature, model_name=llm_model)
+    memory = ConversationBufferMemory()
+    retriever = vector_db.as_retriever()
+    qa_chain = ConversationalRetrievalChain.from_llm(
+        llm=llm,
+        retriever=retriever,
+        memory=memory,
+        return_source_documents=True,
+    )
+    return qa_chain
+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
 
-import chromadb
-import re
-import uuid 
 
-enable_box = gr.Textbox.update(value = None, placeholder = 'Upload your OpenAI API key',interactive = True)
-disable_box = gr.Textbox.update(value = 'OpenAI API key is Set', interactive = False)
+# 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 set_apikey(api_key: str):
-        app.OPENAI_API_KEY = api_key        
-        return disable_box
-    
-def enable_api_box():
-        return enable_box
-
-def add_text(history, text: str):
-    if not text:
-         raise gr.Error('enter text')
-    history = history + [(text,'')] 
-    return history
-
-class my_app:
-    def __init__(self, OPENAI_API_KEY: str = None ) -> None:
-        self.OPENAI_API_KEY: str = OPENAI_API_KEY
-        self.chain = None
-        self.chat_history: list = []
-        self.N: int = 0
-        self.count: int = 0
-
-    def __call__(self, file: str) -> Any:
-        if self.count==0:
-            self.chain = self.build_chain(file)
-            self.count+=1
-        return self.chain
-    
-    def chroma_client(self):
-        #create a chroma client
-        client = chromadb.Client()
-        #create a collecyion
-        collection = client.get_or_create_collection(name="my-collection")
-        return client
+
+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 process_file(self,file: str):
-        loader = PyPDFLoader(file.name)
-        documents = loader.load()  
-        pattern = r"/([^/]+)$"
-        match = re.search(pattern, file.name)
-        file_name = match.group(1)
-        return documents, file_name
+
+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)
     
-    def build_chain(self, file: str):
-        documents, file_name = self.process_file(file)
-        #Load embeddings model
-        embeddings = OpenAIEmbeddings(openai_api_key=self.OPENAI_API_KEY) 
-        pdfsearch = Chroma.from_documents(documents, embeddings, collection_name= file_name,)
-        chain = ConversationalRetrievalChain.from_llm(
-                ChatOpenAI(temperature=0.0, openai_api_key=self.OPENAI_API_KEY), 
-                retriever=pdfsearch.as_retriever(search_kwargs={"k": 1}),
-                return_source_documents=True,)
-        return chain
+    # 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_response(history, query, file): 
-        if not file:
-            raise gr.Error(message='Upload a PDF')    
-        chain = app(file)
-        result = chain({"question": query, 'chat_history':app.chat_history},return_only_outputs=True)
-        app.chat_history += [(query, result["answer"])]
-        app.N = list(result['source_documents'][0])[1][1]['page']
-        for char in result['answer']:
-           history[-1][-1] += char
-           yield history,''
-
-def render_file(file):
-        doc = fitz.open(file.name)
-        page = doc[app.N]
-        #Render the page as a PNG image with a resolution of 300 DPI
-        pix = page.get_pixmap(matrix=fitz.Matrix(300/72, 300/72))
-        image = Image.frombytes('RGB', [pix.width, pix.height], pix.samples)
-        return image
-
-def render_first(file):
-        doc = fitz.open(file.name)
-        page = doc[0]
-        #Render the page as a PNG image with a resolution of 300 DPI
-        pix = page.get_pixmap(matrix=fitz.Matrix(300/72, 300/72))
-        image = Image.frombytes('RGB', [pix.width, pix.height], pix.samples)
-        return image,[]
-
-app = my_app()
-with gr.Blocks() as demo:
-    with gr.Column():
-        with gr.Row():
-            with gr.Column(scale=0.8):
-                api_key = gr.Textbox(placeholder='Enter OpenAI API key', show_label=False, interactive=True).style(container=False)
-            with gr.Column(scale=0.2):
-                change_api_key = gr.Button('Change Key')
-        with gr.Row():           
-            chatbot = gr.Chatbot(value=[], elem_id='chatbot').style(height=650)
-            show_img = gr.Image(label='Upload PDF', tool='select' ).style(height=680)
-    with gr.Row():
-        with gr.Column(scale=0.60):
-            txt = gr.Textbox(
-                        show_label=False,
-                        placeholder="Enter text and press enter",
-                    ).style(container=False)
-        with gr.Column(scale=0.20):
-            submit_btn = gr.Button('submit')
-        with gr.Column(scale=0.20):
-            btn = gr.UploadButton("📁 upload a PDF", file_types=[".pdf"]).style()
+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
+
+
+def demo():
+    with gr.Blocks(theme="base") as demo:
+        vector_db = gr.State()
+        qa_chain = gr.State()
+        collection_name = gr.State()
         
-    api_key.submit(
-            fn=set_apikey, 
-            inputs=[api_key], 
-            outputs=[api_key,])
-    change_api_key.click(
-            fn= enable_api_box,
-            outputs=[api_key])
-    btn.upload(
-            fn=render_first, 
-            inputs=[btn], 
-            outputs=[show_img,chatbot],)
-    
-    submit_btn.click(
-            fn=add_text, 
-            inputs=[chatbot,txt], 
-            outputs=[chatbot, ], 
-            queue=False).success(
-            fn=get_response,
-            inputs = [chatbot, txt, btn],
-            outputs = [chatbot,txt]).success(
-            fn=render_file,
-            inputs = [btn], 
-            outputs=[show_img]
-    )
+        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...")
 
-    
-demo.queue()
-demo.launch()  
+        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)
+
+        # 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)
+
+
+if __name__ == "__main__":
+    api_key = get_openai_api_key()  # Get the API key from the user
+    demo(api_key)  # Pass the API key to the demo function