1.0.0 • Published 7 months ago

palm-conversation-toolkit v1.0.0

Weekly downloads
-
License
ISC
Repository
-
Last release
7 months ago

PaLM Conversation Toolkit

Overview

The PaLM Conversation Toolkit is a Node.js library that simplifies interaction with the Generative Language models from Google AI. It wraps around the DiscussServiceClient to provide a more convenient and easier way to generate replies in a conversational context.

Installation

Before using the PaLM Conversation Toolkit, make sure to install the necessary dependencies.

npm install palm-conversation-toolkit @google-ai/generativelanguage google-auth-library dotenv

Setup

First, create a .env file in your root directory and add your API key.

PALM_API_KEY=your_api_key_here

Usage

To get started with the PaLM Conversation Toolkit, import it into your main file:

const conversationToolkit = require("./palm-conversation-toolkit");

Here is an example of how to initialize and use the toolkit:

const myConversation = new conversationToolkit(client, modelDetails, context, examples);

myConversation.generateReply("What are some cowboy movies?")
    .then(reply => console.log(reply))
    .catch(err => console.log(err));

Parameters

  • client: An instance of DiscussServiceClient from Google AI.
  • modelDetails: An object containing the details of the model you want to use.
    • model: The model name (e.g., "models/chat-bison-001")
    • temperature: Model randomness
    • candidateCount: Number of candidates to consider
    • top_k: Top K sampling value
    • top_p: Top P sampling value
  • context: A string that sets the context for the conversation.
  • examples: An array of example objects that include both an "input" and an "output" field. These examples provide context to guide the model's responses.

Methods

generateReply(inputMessage: string) -> Promise<string>

Generates a reply for the given input message.

  • inputMessage: The user's message that you want the model to respond to.

Error Handling

The method will return null and print an error message to the console if an exception occurs.

Example

For a full example, refer to the index.js file in the repository.

License

This project is licensed under the MIT License.