PaLM_with_Retrieval.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "provenance": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "name": "python"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/manjunath7472/js-xlsx/blob/master/PaLM_with_Retrieval.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "# Embeddings and Retrieval with PaLM and Chroma\n",
        "![palm+chroma.png](data:image/png;base64,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)\n",
        "\n",
        "This notebook demonstrates how easy it is to use the PaLM API, together with Chroma, to build a question answering assistant using information from a pre-prepared set of documents, which cites its sources.\n",
        "\n",
        "We'll use the text of the 2022 and 2023 U.S. state of the union addresses as an example, but you can use any text files you like with a few easy modifications.\n",
        "\n",
        "## How it works\n",
        "\n",
        "This type of application is called 'retrieval augmented generation', becasuse we use both a retrieval system (Chroma), and a generative model (PaLM) to achieve the result.\n",
        "\n",
        "The basic flow is as follows:\n",
        "\n",
        "0. The text documents are loaded line by line, then embedded using the PaLM embeddings API and stored in a Chroma collection. You can learn more about Embeddings and how they work [here](https://developers.generativeai.google/tutorials/embeddings_quickstart#what_are_embeddings).\n",
        "\n",
        "1. When the user submits a query, it gets embedded using the same model as the documents, and the lines most relevant to the query are retrieved by Chroma.\n",
        "\n",
        "2. The user-submitted question is passed to the PaLM text completion API, along with the extra context retrieved by Chroma. The PaLM API then generates a response based on the source context.\n",
        "\n",
        "3. The response is displayed to the user, along with the source lines used as extra context."
      ],
      "metadata": {
        "id": "5JNBGiogRM43"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Setup\n"
      ],
      "metadata": {
        "id": "XAKdtI0yTq26"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "We'll install the PaLM and Chroma dependencies - just two packages.\n",
        "We also import some convenient libraries for working with files and showing progress bars."
      ],
      "metadata": {
        "id": "PzuJe0-0UzIy"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "R1nkKfEUepiH"
      },
      "outputs": [],
      "source": [
        "%pip install -Uq chromadb google-generativeai\n",
        "\n",
        "import chromadb\n",
        "import google.generativeai as palm\n",
        "import glob\n",
        "from tqdm import tqdm"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "\n",
        "You will need a PaLM API key to run this example. You can get one [here](https://developers.generativeai.google/tutorials/setup?authuser=1)."
      ],
      "metadata": {
        "id": "DHN8I1QDUxfV"
      }
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "w_s0YNK7AYiv"
      },
      "outputs": [],
      "source": [
        "api_key = \"YOUR_API_KEY\"\n",
        "palm.configure(api_key=api_key)"
      ]
    },
    {
      "cell_type": "markdown",
      "source": [
        "We'll download the example documents from the [Chroma repository](https://github.com/chroma-core/chroma) on GitHub."
      ],
      "metadata": {
        "id": "7C1oYt_3UaGT"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "!wget https://raw.githubusercontent.com/chroma-core/chroma/main/examples/chat_with_your_documents/documents/state_of_the_union_2022.txt\n",
        "!wget https://raw.githubusercontent.com/chroma-core/chroma/main/examples/chat_with_your_documents/documents/state_of_the_union_2023.txt"
      ],
      "metadata": {
        "id": "DqR9ckGmmesa"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Loading and ingesting data\n",
        "\n",
        "We load the text data into Chroma. The documents will be embedded using the PaLM embedding function, which comes built in with Chroma."
      ],
      "metadata": {
        "id": "REGmvLcuWLoX"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "# First, create the PaLM embedding function\n",
        "from chromadb.utils.embedding_functions import GooglePalmEmbeddingFunction\n",
        "palm_embedding = GooglePalmEmbeddingFunction(api_key=api_key)\n",
        "\n",
        "# Create an Ephemeral Chroma client (good for development and experimentation)\n",
        "client = chromadb.EphemeralClient()\n",
        "# Create a Chroma collection which uses the PaLM embedding function\n",
        "collection = client.create_collection(name='state_of_the_union', embedding_function=palm_embedding)\n",
        "\n",
        "# Get the names of the downloaded text files.\n",
        "text_files = glob.glob('*.txt')\n",
        "\n",
        "count = 0\n",
        "for filename in text_files:\n",
        "  with open(filename, 'r') as file:\n",
        "    for line_number, line in enumerate(\n",
        "        tqdm((file.readlines()), desc=f\"Reading {filename}\")\n",
        "    ):\n",
        "      line = line.strip()\n",
        "      # Skip empty lines\n",
        "      if line == '':\n",
        "        continue\n",
        "      collection.add(\n",
        "          ids=[str(count)], # We use simple incremental IDs, but these can be any string\n",
        "          documents=[line], # Each line that contains text is stored as an individual document in Chroma\n",
        "          metadatas=[{'filename': filename, 'line_number': line_number}] # We store the source of the embedded line as metadata\n",
        "      )\n",
        "      count += 1\n"
      ],
      "metadata": {
        "id": "-yvHlg9FoytS"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Text completion with context\n",
        "\n",
        "We construct a prompt which tells the PaLM text completion model to answer queries on the basis of information available in the documents."
      ],
      "metadata": {
        "id": "WUjaNz0Ead5a"
      }
    },
    {
      "cell_type": "markdown",
      "source": [
        "Building a prompt that includes retrieved context."
      ],
      "metadata": {
        "id": "AzONwdKAeomw"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "from typing import List\n",
        "def build_prompt(query: str, context: List[str]):\n",
        "  joined_context = \"\\n\".join(context)\n",
        "\n",
        "  return f\"\"\"\n",
        "You are a helpful assistant tasked with answering queries based on information provided to you.\n",
        "\n",
        "Please answer the following query.\n",
        "\n",
        "{query}\n",
        "\n",
        "Using all of the following information, and discuss any relevant information point by point.\n",
        "\n",
        "{joined_context}\n",
        "\n",
        "----------------\n",
        "\n",
        "Important: Only use the information provided.\n",
        "If there isn't enough information provided to answer the query, say so.\n",
        "Make your answer a complete sentence.\n",
        "\"\"\""
      ],
      "metadata": {
        "id": "ZwulGv7yz99C"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Using Chroma to retrieve context relevant to the query for use by the text completion model. We also return the source file and line number for the retrieved context."
      ],
      "metadata": {
        "id": "FvhF2Rkaes2C"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def get_context(query: str):\n",
        "  results = collection.query(query_texts=[query], n_results=3)\n",
        "\n",
        "  context = results['documents'][0]\n",
        "  metadatas = results['metadatas'][0]\n",
        "  # Unpack the metadata into a nicely printable string\n",
        "  sources = [f\"file: {metadata['filename']} line: {metadata['line_number']}\" for metadata in metadatas]\n",
        "\n",
        "  return results['documents'][0], sources"
      ],
      "metadata": {
        "id": "7eM3CNXhFTiF"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Finally, we put it all together in a single function"
      ],
      "metadata": {
        "id": "pyw7Z-Qie6G3"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "def answer_query(query: str):\n",
        "  context, sources = get_context(query=query)\n",
        "  prompt = build_prompt(query=query, context=context)\n",
        "\n",
        "  completion = palm.generate_text(prompt=prompt)\n",
        "  print(completion.result)\n",
        "  print(\"Sources:\")\n",
        "  print('\\n'.join(sources))\n"
      ],
      "metadata": {
        "id": "TrNzUY6LGytf"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Trying it out\n",
        "\n",
        "Here are some example queries with their outputs."
      ],
      "metadata": {
        "id": "LlRgLifMfGON"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "answer_query(\"What was said about the war in Ukraine?\")"
      ],
      "metadata": {
        "id": "d3hyEY45HqzT"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "answer_query(\"What happened during the covid pandemic?\")"
      ],
      "metadata": {
        "id": "u7iAPmR3Kfli"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "source": [
        "answer_query(\"Has there been much investment in American manufacturing?\")"
      ],
      "metadata": {
        "id": "7zTorxZDKfs5"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "Let's try one where there's no information in the documents"
      ],
      "metadata": {
        "id": "YPInAY2pfKKg"
      }
    },
    {
      "cell_type": "code",
      "source": [
        "answer_query(\"Who is Zaphod Beeblebrox?\")"
      ],
      "metadata": {
        "id": "iBCTFSyYfMOK"
      },
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "source": [
        "## Next Steps\n",
        "\n",
        "Check out the [Chroma](https://docs.trychroma.com/getting-started) and [PaLM](https://developers.generativeai.google/guide) documentation to learn more about building your own applications with AI and retrieval.\n",
        "\n",
        "Learn how to [deploy Chroma to GCP](https://github.com/chroma-core/chroma/tree/main/examples/deployments/google-cloud-compute) using Terraform, for production applications.  \n",
        "\n",
        "Join the [Chroma](https://discord.gg/MMeYNTmh3x) and [PaLM](https://discord.gg/google-dev-community) developer communities to learn more and get help with your projects.\n",
        "\n",
        "Chroma is [hiring](https://trychroma.notion.site/careers-chroma-9d017c3007c7478ebd85bad854101497?pvs=4)!"
      ],
      "metadata": {
        "id": "XYQk8paegQVS"
      }
    }
  ]
}