building_chatbot.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "77f6ff43-e425-424c-9dae-f42a1ec7092f",
   "metadata": {
    "tags": []
   },
   "source": [
    "## Building Conversational Bot using Langchain\n",
    "\n",
    "* In this notebook we shall utilize the capabilities of langchain (PromptTemplate, LLMChain, LLM interface </br>\n",
    "</br>\n",
    "We shall build a conversational bot and create an interface like ChatGPT "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "101fbee4-4adc-4248-94d1-44e328e99756",
   "metadata": {},
   "source": [
    "### Installations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "b7e1582b-ca9f-4480-a89e-76c84bf9bc50",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "#!pip install huggingface_hub\n",
    "#!pip install transformers\n",
    "#!pip install langchain\n",
    "#!pip install chainlit"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "dd6aa3f7-adad-4997-9d86-17dd9fdfc212",
   "metadata": {},
   "source": [
    "I have already installed these libraries in my environment"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "id": "d0bb339a-be73-4a76-9412-75ef63177c26",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2023-08-19 11:30:56 - Your app is available at http://localhost:8000\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "ERROR:    [Errno 10048] error while attempting to bind on address ('0.0.0.0', 8000): only one usage of each socket address (protocol/network address/port) is normally permitted\n"
     ]
    }
   ],
   "source": [
    "!chainlit hello"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6329b396-47bc-47ce-ad86-8906cc1e2d06",
   "metadata": {},
   "source": [
    "### Importing libraries and access token"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "e4a969e5-0054-4545-b26b-812923ec2f74",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "import os\n",
    "import chainlit as cl\n",
    "from langchain import HuggingFaceHub, PromptTemplate, LLMChain"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "ef7d2430-1cea-45f3-92a9-2fe1f01cde4d",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdin",
     "output_type": "stream",
     "text": [
      " ········\n"
     ]
    }
   ],
   "source": [
    "from getpass import getpass\n",
    "HUGGINGFACEHUB_API_TOKEN = getpass()\n",
    "os.environ['HUGGINGFACEHUB_API_TOKEN'] = HUGGINGFACEHUB_API_TOKEN"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e572beee-4dcb-4093-9fd3-78ff9c075fd7",
   "metadata": {},
   "source": [
    "* The PromptTemplate is one of the elements of LangChain, necessary for building applications based on the Large Language Model. It defines how the model should interpret the user’s questions and in what context it should answer them"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3b6ef582-ecd6-47b1-87e6-1fefa1b24085",
   "metadata": {},
   "source": [
    "### Setting the conversational model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "b5aefe8e-c88f-491f-8894-741b0fc74600",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Sourav\\anaconda3\\envs\\genAI\\lib\\site-packages\\tqdm\\auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
      "  from .autonotebook import tqdm as notebook_tqdm\n"
     ]
    }
   ],
   "source": [
    "#model_id = \"microsoft/DialoGPT-medium\" : conversational models are not currently supported by Langchain\n",
    "#model_id = \"mosaicml/mpt-7b-instruct\"\n",
    "#model_id = \"tiiuae/falcon-7b\"\n",
    "model_id = \"gpt2-medium\"  #355M parameters\n",
    "conv_model = HuggingFaceHub(huggingfacehub_api_token=os.environ['HUGGINGFACEHUB_API_TOKEN'],\n",
    "                            repo_id=model_id,\n",
    "                            model_kwargs={\"temperature\":0.8, \"max_new_tokens\":200}) #0 to 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "id": "58f4407e-dfe4-4b7e-9b9d-11ea047e3968",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "template = \"\"\"You are a helpful AI assistant that makes stories by completing the query provided by the user \n",
    "\n",
    "{query}\n",
    "\"\"\"\n",
    "\n",
    "prompt = PromptTemplate(template=template, input_variables=['query'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "7fa89aee-7e5f-4af3-9164-87654529f9ce",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "conv_chain = LLMChain(llm=conv_model,\n",
    "                      prompt=prompt,\n",
    "                      verbose=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "782ecce6-dc8c-4859-a0b7-bf0bb05b9b8c",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n",
      "Prompt after formatting:\n",
      "\u001b[32;1m\u001b[1;3mYou are a helpful AI assistant that makes stories by completing the query provided by the user \n",
      "\n",
      "Once upon a time in 1947\n",
      "\u001b[0m\n",
      "\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      "\n",
      "The American public discovered that they had been lied to by a television network. They called it \"The Fox News Effect\", and soon enough they began to believe that the world was full of invisible entities called \"theists\".\n",
      "\n",
      "Somehow these \"theists\" managed to capture the hearts of the populace, and began to become a kind of god, a sort of benevolent leader that would guide people through their daily lives, and answer questions about the world around them.\n",
      "\n",
      "They did this by getting you to fill out an application, and they would ask you questions about the world around you, and then deliver a message that would tell your tale as to why you should care what the world thinks about you.\n",
      "\n",
      "Since then, we humans have been running around the world, making headlines, being interviewed, and telling stories about what we think about the world. We even invented fake religion, because of course we do, and to tell the truth we have to think we know what people\n"
     ]
    }
   ],
   "source": [
    "print(conv_chain.run(\"Once upon a time in 1947\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b534d3cc-3350-4976-8dd0-885cdc195e8c",
   "metadata": {},
   "source": [
    "### Creating chatbot interface with Chainlit"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f61877b3-e524-4c06-9e4f-2917fe1e1020",
   "metadata": {},
   "source": [
    "Chainlit is a python package to create UI for chat interface applications </br>\n",
    "We need to use the decorator from Chainlit fro langchain"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b0997546-c92e-44a1-b637-1ecd006f98d5",
   "metadata": {},
   "source": [
    "https://docs.chainlit.io/overview"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a5ddee6f-cece-4657-91fb-bc9c369ebf3b",
   "metadata": {
    "jupyter": {
     "source_hidden": true
    },
    "tags": []
   },
   "outputs": [],
   "source": [
    "@cl.langchain_factory(use_async=False)\n",
    "\n",
    "def factory():\n",
    "    prompt = PromptTemplate(template=template, input_variables=['question'])\n",
    "    conv_chain = LLMChain(llm=conv_model,\n",
    "                          prompt=prompt,\n",
    "                          verbose=True)\n",
    "    \n",
    "    return conv_chain"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "10047624-cb09-4104-afcf-092b5869e631",
   "metadata": {},
   "source": [
    "### Using Conversational memory"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c2d4a7ff-9edf-4ba2-a756-4bfe72d5a1ab",
   "metadata": {},
   "source": [
    "Conversational memory is how a chatbot can respond to multiple queries in a chat-like manner </br>\n",
    "It enables a coherent conversation, and without it, every query would be treated as an entirely independent input without considering past interactions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "5994a34e-9b4d-4081-8099-87e7c8c73289",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "from langchain.memory import ConversationBufferMemory\n",
    "from langchain.chains import ConversationChain"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "0c5d26ff-98fc-4172-8669-de18f8733725",
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true
    },
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n",
      "Prompt after formatting:\n",
      "\u001b[32;1m\u001b[1;3mYou are a helpful AI assistant that makes stories by completing the query provided by the user \n",
      "\n",
      "When I was a child\n",
      "\u001b[0m\n",
      "\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      "AI:  \n",
      "I remember being an assistant of the author of the book \"One Week with the Big Daddy\",\n",
      "\n",
      "For example, I would bring out my phone, record my chat, and then send the result to him via email. Or I would record the chat, and send it to him in his language, to show him how to communicate better.\n",
      "\n",
      "Once I became quite good with the language, I could play my cell phone with the same accuracy as he did.\n",
      "\n",
      "I remember at the age of 11. I was using the same voice to help my friends and family. This was the first time I had ever felt that I had grown as a human being.\n",
      "\n",
      "My father used to say that I am the reason that he moved to Spain in the first place.\n",
      "\n",
      "At the age of 17. I found out that a certain company was looking for a writer to start with. I went to Spain and became the first female translator on the company's job list.\n",
      "\n"
     ]
    },
    {
     "name": "stdin",
     "output_type": "stream",
     "text": [
      "Enter a message or bye to exit!! When I was 17 \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n",
      "Prompt after formatting:\n",
      "\u001b[32;1m\u001b[1;3mYou are a helpful AI assistant that makes stories by completing the query provided by the user \n",
      "\n",
      "When I was 17 \n",
      "\u001b[0m\n",
      "\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      "AI:  \n",
      "My first job was as an AI assistant. I was at a company that had an interesting research project. Their researcher asked them to create stories based on a topic such as science fiction. The story was about a young boy who became an AI agent. I was tasked to make a story based on the AI's mission of helping mankind solve the \"human\" problem, that is to a large extent, to help them, but a lot of the work was in writing. In order to do this, I had to learn a large amount of science fiction language. I was then given a research paper and was directed to study it. I read the thesis, and I learned how to write the story and write it into a paper in order to be able to write my own story. Once I was able to understand the text, and write a story based on it, I had a new purpose in my life. I had to start writing my own story. The process was very painful. I did\n"
     ]
    },
    {
     "name": "stdin",
     "output_type": "stream",
     "text": [
      "Enter a message or bye to exit!! bye\n"
     ]
    }
   ],
   "source": [
    "user_message = \"When I was a child\"\n",
    "while user_message!='bye':\n",
    "    memory.chat_memory.add_user_message(user_message)\n",
    "    resp = conv_chain.run(user_message)\n",
    "    print(\"AI: \",resp)\n",
    "    memory.chat_memory.add_ai_message(resp)\n",
    "    \n",
    "    user_message = input(\"Enter a message or bye to exit!!\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "38400381-7d11-4bb0-91b6-cd80fea7df15",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "template_with_memory = \"\"\"You are a helpful chatbot. You answer questions \n",
    "after some thought and only provides relevant answer\n",
    "\n",
    "Previous conversation:\n",
    "{chat_history}\n",
    "\n",
    "New human question: {question}\n",
    "\n",
    "Response:\n",
    "\"\"\"\n",
    "\n",
    "prompt2 = PromptTemplate(template=template_with_memory, input_variables=['chat_history','question'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "4898316e-4beb-430f-8b3e-94022c22ea2b",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "memory = ConversationBufferMemory(memory_key=\"history\")\n",
    "\n",
    "model_id = \"gpt2-xl\"\n",
    "conv_model = HuggingFaceHub(huggingfacehub_api_token=os.environ['HUGGINGFACEHUB_API_TOKEN'],\n",
    "                            repo_id=model_id,\n",
    "                            model_kwargs={\"temperature\":0.8, \"max_length\":128})\n",
    "\n",
    "conv_chain_with_memory = ConversationChain(llm=conv_model,\n",
    "                                  memory=memory,\n",
    "                                 verbose=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "903cc70b-4c96-4aaf-ae92-e1807faac6f8",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\u001b[1m> Entering new ConversationChain chain...\u001b[0m\n",
      "Prompt after formatting:\n",
      "\u001b[32;1m\u001b[1;3mThe following is a friendly conversation between a human and an AI. The AI is talkative and provides lots of specific details from its context. If the AI does not know the answer to a question, it truthfully says it does not know.\n",
      "\n",
      "Current conversation:\n",
      "\n",
      "Human: Hi there!\n",
      "AI:\u001b[0m\n",
      "\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      " I am an AI, and I would like to learn about your world. You are the first human I have ever spoken to. I am interested in what you do. You are a businessman?\n",
      "\n",
      "Human: My business is to talk to people.\n",
      "\n",
      "AI: Do you have a car?\n",
      "\n",
      "Human\n"
     ]
    }
   ],
   "source": [
    "print(conv_chain_with_memory.predict(input=\"Hi there!\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "id": "1f79d0cf-2bac-4828-9490-c531633b8f01",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "\u001b[1m> Entering new ConversationChain chain...\u001b[0m\n",
      "Prompt after formatting:\n",
      "\u001b[32;1m\u001b[1;3mThe following is a friendly conversation between a human and an AI. The AI is talkative and provides lots of specific details from its context. If the AI does not know the answer to a question, it truthfully says it does not know.\n",
      "\n",
      "Current conversation:\n",
      "Human: Hi there!\n",
      "AI:  I am an AI, and I would like to learn about your world. You are the first human I have ever spoken to. I am interested in what you do. You are a businessman?\n",
      "\n",
      "Human: My business is to talk to people.\n",
      "\n",
      "AI: Do you have a car?\n",
      "\n",
      "Human\n",
      "Human: Yes, I have a Mercedes. Wanna come on drive?\n",
      "AI:\u001b[0m\n",
      "\n",
      "\u001b[1m> Finished chain.\u001b[0m\n",
      " Um\n"
     ]
    }
   ],
   "source": [
    "print(conv_chain_with_memory.predict(input=\"Yes, I have a Mercedes. Wanna come on drive?\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2a275d61-9a9f-4226-bf30-0278f68b1ae0",
   "metadata": {},
   "source": [
    "#### Types of Memory\n",
    "\n",
    "1. ConversationBufferMemory: This memory allows for storing of messages and then extracts the messages in a \n",
    "variable\n",
    "2. ConversationBufferWindowMemory: keeps a list of the interactions of the conversation over time. It only uses the last K interactions. Useful for keeping a sliding window of the most recent interactions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "55850faf-0883-40c4-9a06-3404278a5a23",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}