Together AI Dart Package

The Together AI Dart package provides a convenient way to interact with the Together AI API using Dart. It offers a simple and intuitive interface for generating text, images, and more using various AI models.

Features

• Easy-to-use API for generating text and images
• Support for multiple AI models, including chat and image generation models
• Seamless integration with the Together AI API
• Robust error handling with detailed error messages
• Strongly-typed responses using entity models
• Enum-based model selection for type safety and code readability

Installation

To use the Together AI Dart package in your project, add the following dependency to your pubspec.yaml file:

dependencies:
  together_ai_sdk: ^2.0.2

Then, run dart pub get to fetch the package.

Usage

First, import the package in your Dart code:

import 'package:together_ai_sdk/together_ai_sdk.dart';

Security Disclaimer

⚠️ IMPORTANT: API Key Security ⚠️

When using this SDK, please be aware of the following security considerations:

1. Never expose your API key in client-side code. The TogetherAISdK initialization with an API key should not be done directly in mobile or web applications that are distributed to end-users.

2. Use environment variables to store your API key during development and testing. Never commit these to version control.

Initializing the Together AI SDK

Create an instance of the TogetherAISDK class with your API key and the desired model:

final togetherAI = TogetherAISdK('YOUR_API_KEY');

Generating Chat

To generate text using a chat model, use the chatCompletion method:

  final chatResponse = await togetherAISdk.chatCompletion([
    {'role': 'system', 'content': 'You are a helpful'},
    {'role': 'user', 'content': 'Once upon a'},
  ], ChatModel.qwen15Chat72B);

  print(chatResponse);

Generating Chat using stream

To generate a stream, use the streamResponse method:

  final chatResponse = await togetherAISdk.streamResponse([
    {'role': 'system', 'content': 'You are a helpful AI assistant'},
    {'role': 'user', 'content': 'Tell me a cyberpunk story'},
  ], ChatModel.qwen15Chat72B);

  print(chatResponse);

Generating Chat With Vision Models

To generate text using a vision model, use the visionChatCompletion method:

  String imagePath =
      '/Users/kimchi/Documents/Packages/together_ai_sdk/example/your_image_name.jpeg';

  //This function converts the image to base64
  String base64ImageUrl = await imageToBase64(imagePath);

//The message object is a list of maps with role and content keys. Which is not the same as the Message class in the SDK.
  final messages = [
    {'role': 'system', 'content': 'You are a helpful AI'},
    {
      'role': 'user',
      'content': [
        {'type': 'text', 'text': 'What\'s in this image?'},
        {
          'type': 'image_url',
          'image_url': {'url': base64ImageUrl}
        }
      ]
    },
  ];

  //Please use ChatVision model.
  //LlamaVFree, the free model is included.

  final chatVisionResponse = await togetherAISdk.visionChatCompletion(
      messages, ChatVisionModel.llama3290BV);

  print(chatVisionResponse);

Generating Text

To generate text, use the textCompletion method:

 final textResponse = await togetherAISdk.textCompletion(
      'Once upon a', LanguageModel.qwen1572B);

  print(textResponse);

Simple conversation memory class

To use simple memory, use the ConversationBufferWindowMemory class or ConversationBufferMemory:

  ConversationMemory memory = ConversationBufferWindowMemory(4);
  memory.addMessage(chatResponse.choices[0].message);

  print(memory.getMessages());

  //Printing that gives you 
  //[{role: assistant, content: Diverse, multicultural, modern, cosmopolitan, city-state, financial hub, green spaces, food paradise, vibrant nightlife,狮城。}]

Generating Images

To generate an image, use the imageGeneration method:

  final imageResponse = await togetherAISdk.imageGeneration('Group of cats', imageModel: ImageModel.stableDiffusionXL1_0);

  print(imageResponse);

Please note currently there is only one model that is supported.

1. stabilityai/stable-diffusion-xl-base-1.0

Error Handling

The Together AI Dart package provides comprehensive error handling. If an error occurs during an API request, a TogetherAIError instance will be thrown. You can catch and handle specific error types based on the error code.

try {
  final response = await togetherAI.chatCompletion(...);
  // Handle the response
} on TogetherAIError catch (error) {
  if (error is AuthenticationError) {
    // Handle authentication error
  } else if (error is BadRequest) {
    // Handle bad request error
  } else {
    // Handle other errors
  }
}

Models

The Together AI Dart package supports various chat and image generation models. You can find the available models in the ChatModel, ImageModel and ChatVisionModel enums, respectively in the enum_ai_models.dart file, which is in the common folder. If there are any missing models please let me know.

Conflicts

If you have any conflicts with your existing codebase, make sure to import the sdk using the as keyword.

Contributing

Contributions to the Together AI Dart package are welcome! If you find any issues or have suggestions for improvements, please open an issue or submit a pull request on the GitHub repository.

Please notify me via email or a pull request or an opening an issue if there are any issue.

License

This package is released under the MIT License.