app.py

import gradio as gr
import openai
import requests
import csv
import os
import langchain
import chromadb
import glob
import pickle

from PyPDF2 import PdfReader
from PyPDF2 import PdfWriter
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.text_splitter import CharacterTextSplitter
#from langchain.vectorstores import ElasticVectorSearch, Pinecone, Weaviate, FAISS
from langchain.chains.question_answering import load_qa_chain
from langchain.llms import OpenAI

from langchain.embeddings.openai import OpenAIEmbeddings
#from langchain.vectorstores import Chroma
#from langchain.text_splitter import TokenTextSplitter
#from langchain.llms import OpenAI
from langchain import OpenAI
from langchain.chat_models import ChatOpenAI
#from langchain.chains import ChatVectorDBChain
#from langchain.chains import RetrievalQA
from langchain.document_loaders import PyPDFLoader
from langchain.chains.question_answering import load_qa_chain

# Use Chroma in Colab to create vector embeddings, I then saved them to HuggingFace so now I have to set it use them here.
#from chromadb.config import Settings
#client = chromadb.Client(Settings(
##    chroma_db_impl="duckdb+parquet",
#    persist_directory="./embeddings" # Optional, defaults to .chromadb/ in the current directory
#))






def get_empty_state():
    return {"total_tokens": 0, "messages": []}

    
#Initial prompt template, others added below from TXT file
prompt_templates = {"All Needs Experts": "Respond as if you are combiation of all needs assessment experts."}
actor_description = {}

def download_prompt_templates():
    url = "https://huggingface.co/spaces/ryanrwatkins/needs/raw/main/gurus.txt"
    try:
        response = requests.get(url)
        reader = csv.reader(response.text.splitlines())
        next(reader)  # skip the header row
        for row in reader:
            if len(row) >= 2:
                act = row[0].strip('"')
                prompt = row[1].strip('"')
                description = row[2].strip('"')
                prompt_templates[act] = prompt
                actor_description[act] = description
                


    except requests.exceptions.RequestException as e:
        print(f"An error occurred while downloading prompt templates: {e}")
        return

    choices = list(prompt_templates.keys())
    choices = choices[:1] + sorted(choices[1:])
    return gr.update(value=choices[0], choices=choices)

def on_prompt_template_change(prompt_template):
    if not isinstance(prompt_template, str): return
    return prompt_templates[prompt_template]

def on_prompt_template_change_description(prompt_template):
    if not isinstance(prompt_template, str): return
    return actor_description[prompt_template]




def submit_message(prompt, prompt_template, temperature, max_tokens, context_length, state):

    openai.api_key = os.environ['openai_key']
    os.environ["OPENAI_API_KEY"] = os.environ['openai_key']
    
    # load in all the files
    #path = './files'
    #pdf_files = glob.glob(os.path.join(path, "*.pdf"))
    #pdf_files = glob.glob(os.path.join(path, "*.pdf"))

    #for file in pdf_files:
     # loader = PyPDFLoader(file)
     # pages = loader.load_and_split()
     # text_splitter = TokenTextSplitter(chunk_size=1000, chunk_overlap=0)
     # split_pages = text_splitter.split_documents(pages)
     
    #persist_directory = "./embeddings"
    #embeddings = OpenAIEmbeddings()
    #vectordb = Chroma.from_documents(split_pages, embeddings, persist_directory=persist_directory)
    #vectordb.persist()

    path = './files'
    pdf_files = glob.glob(os.path.join(path, "*.pdf"))
    
    merger = PdfWriter()
    
    # add all file in the list to the merger object
    for pdf in pdf_files:
      merger.append(pdf)
    merger.write("merged-pdf.pdf")
    merger.close()
    
    reader = PdfReader("merged-pdf.pdf")
    raw_text = ''
    for i, page in enumerate(reader.pages):
      text = page.extract_text()
      if text:
          raw_text += text
    text_splitter = CharacterTextSplitter(        
      separator = "\n",
      chunk_size = 1000,
      chunk_overlap  = 200,
      length_function = len,
    )
    texts = text_splitter.split_text(raw_text)
    len(texts)
    embeddings = OpenAIEmbeddings()
        
    
    history = state['messages']

    if not prompt:
        return gr.update(value=''), [(history[i]['content'], history[i+1]['content']) for i in range(0, len(history)-1, 2)], f"Total tokens used: {state['total_tokens']}", state
    
    prompt_template = prompt_templates[prompt_template]

    system_prompt = []
    if prompt_template:
        system_prompt = [{ "role": "system", "content": prompt_template }]

    prompt_msg = { "role": "user", "content": prompt }


    try:
        #completion = openai.ChatCompletion.create(model="gpt-3.5-turbo", messages=system_prompt + history[-context_length*2:] + [prompt_msg], temperature=temperature, max_tokens=max_tokens)

# completion = openai.ChatCompletion.create(model="gpt-3.5-turbo", messages=system_prompt + history[-context_length*2:] + [prompt_msg], temperature=temperature, max_tokens=max_tokens)

        #completion_chain = load_qa_chain(ChatOpenAI(temperature=temperature, max_tokens=max_tokens, model_name="gpt-3.5-turbo"), chain_type="stuff" )
        #completion = RetrievalQA(combine_documents_chain=completion_chain, retriever=vectordb.as_retriever())
        #query = str(system_prompt + history[-context_length*2:] +  [prompt_msg])
        #completion = completion.run(query)
        # from https://blog.devgenius.io/chat-with-document-s-using-openai-chatgpt-api-and-text-embedding-6a0ce3dc8bc8
        #completion_chain = load_qa_chain(ChatOpenAI(temperature=temperature, max_tokens=max_tokens, model_name="gpt-3.5-turbo"), chain_type="stuff" )
        #completion = RetrievalQA(combine_documents_chain=completion_chain, retriever=vectordb.as_retriever(), return_source_documents=False)
        #completion = RetrievalQA.from_chain_type(llm=ChatOpenAI(temperature=temperature, max_tokens=max_tokens, model_name="gpt-3.5-turbo"), chain_type="stuff", retriever=vectordb.as_retriever(), return_source_documents=True)
        #query = str(system_prompt + history[-context_length*2:] +  [prompt_msg])
        #completion = completion({"query": query})
        #completion = completion.run(query)

#        completion = completion({"question": query, "chat_history": history[-context_length*2:]})

        #with open("foo.pkl", 'rb') as f:
        #    new_docsearch = pickle.load(f)
        
        docsearch = FAISS.from_texts(texts, embeddings)
        #query = str(system_prompt + history[-context_length*2:] +  [prompt_msg])
        query = str(system_prompt + history +  [prompt_msg])
        docs = docsearch.similarity_search(query)
        #print(docs[0].page_content)

        chain = load_qa_chain(ChatOpenAI(temperature=temperature, max_tokens=max_tokens, model_name="gpt-3.5-turbo"), chain_type="stuff")
        completion = chain.run(input_documents=docs, question=query)
        completion = { "content": completion }
        
# VectorDBQA.from_chain_type(llm=OpenAI(), chain_type="stuff", vectorstore=docsearch, return_source_documents=True)
# https://colab.research.google.com/drive/1dzdNDZyofRB0f2KIB4gHXmIza7ehMX30?usp=sharing#scrollTo=b-ejDn_JfpWW

        get_empty_state()
        state.append(completion.copy())
        
        #history.append(prompt_msg.copy())
        #history.append(completion.copy())
        #history.append(completion.choices[0].message.to_dict())
        #history.append(completion["result"].choices[0].message.to_dict())

        state['total_tokens'] += completion['usage']['total_tokens']
    
    except Exception as e:
        history.append(prompt_msg.copy())
        error  = {
            "role": "system",
            "content": f"Error: {e}"
        }
        history.append(error.copy())

    total_tokens_used_msg = f"Total tokens used: {state['total_tokens']}"

    
    
    chat_messages = [(prompt_msg['content'], completion['content'])]
    #chat_messages = [(history[i]['content'], history[i+1]['content']) for i in range(0, len(history)-1, 2)]
    #chat_messages = [(history[-2]['content'], history[-1]['content'])]
    
    return '', chat_messages, total_tokens_used_msg, state

def clear_conversation():
    return gr.update(value=None, visible=True), None, "", get_empty_state()


css = """
      #col-container {max-width: 80%; margin-left: auto; margin-right: auto;}
      #chatbox {min-height: 400px;}
      #header {text-align: center;}
      #prompt_template_preview {padding: 1em; border-width: 1px; border-style: solid; border-color: #e0e0e0; border-radius: 4px;}
      #total_tokens_str {text-align: right; font-size: 0.8em; color: #666;}
      #label {font-size: 0.8em; padding: 0.5em; margin: 0;}
      .message { font-size: 1.2em; }
      """

with gr.Blocks(css=css) as demo:
    
    state = gr.State(get_empty_state())


    with gr.Column(elem_id="col-container"):
        with open("embeddings.pkl", 'rb') as f: 
            new_docsearch = pickle.load(f)
            
        query = str("performance")
        docs = new_docsearch.similarity_search(query)
           
        gr.Markdown("""# Chat with Needs Assessment Experts (Past and Present)
                    ## Ask questions of experts on needs assessments, get responses from *needs assessment* version of ChatGPT.
                    Ask questions of all of them, or pick your expert.""" + docs,
                    elem_id="header")
        
 
        
        
        with gr.Row():
            with gr.Column():
                chatbot = gr.Chatbot(elem_id="chatbox")
                input_message = gr.Textbox(show_label=False, placeholder="Enter your needs assessment question and press enter", visible=True).style(container=False)
                btn_submit = gr.Button("Submit")
                total_tokens_str = gr.Markdown(elem_id="total_tokens_str")
                btn_clear_conversation = gr.Button("Start New Conversation")
            with gr.Column():
                prompt_template = gr.Dropdown(label="Choose a expert:", choices=list(prompt_templates.keys()))
                prompt_template_preview = gr.Markdown(elem_id="prompt_template_preview")
                with gr.Accordion("Advanced parameters", open=False):
                    temperature = gr.Slider(minimum=0, maximum=2.0, value=0.7, step=0.1, label="Flexibility", info="Higher = more creative/chaotic, Lower = just the guru")
                    max_tokens = gr.Slider(minimum=100, maximum=400, value=200, step=1, label="Max tokens per response")
                    context_length = gr.Slider(minimum=1, maximum=5, value=2, step=1, label="Context length", info="Number of previous questions you have asked. Be careful with high values, it can blow up the token budget quickly.")

   
    btn_submit.click(submit_message, [ input_message, prompt_template, temperature, max_tokens, context_length, state], [input_message, chatbot, total_tokens_str, state])
    input_message.submit(submit_message, [ input_message, prompt_template, temperature, max_tokens, context_length, state], [input_message, chatbot, total_tokens_str, state])
    btn_clear_conversation.click(clear_conversation, [], [input_message, chatbot, total_tokens_str, state])
    #prompt_template.change(on_prompt_template_change, inputs=[prompt_template], outputs=[prompt_template_preview])
    prompt_template.change(on_prompt_template_change, inputs=[prompt_template], outputs=[prompt_template_preview])

    
    demo.load(download_prompt_templates, inputs=None, outputs=[prompt_template], queur=False)


demo.queue(concurrency_count=10)
demo.launch(height='800px')