generate - Access Generative AI Models Seamlessly with Python

Introduction to Generate

Generate is an innovative Python package designed to facilitate access to generative models from multiple platforms through a single unified API. It caters to developers and data scientists who need seamless integration with various AI and machine learning models, enhancing flexibility and productivity.

Overview

Generate simplifies the process of interacting with numerous generative platforms. Whether you're working with OpenAI, Azure, Minimax, or several others, Generate provides a consistent interface that abstracts away platform-specific complexities. This toolset enables users to focus more on their applications' core functionalities rather than the nuances of each individual API.

Features

Multimodal Capabilities: Generate supports a variety of generative tasks, including text, image, and speech generation. It can also handle structured data creation and multimodal text generation, broadening the scope of applications it can support.
Cross-Platform Integration: Users can access over ten platforms with Generate, without needing to worry about differences in message formats, inference parameters, or API structures. This includes both domestic and international platforms.
Unified API: Generate provides a uniform API for accessing different services, which means users only have to learn a single interface to interact with multiple models across different platforms.
Asynchronous and Streaming Calls: The package is designed to accommodate various calling patterns, including synchronous, asynchronous, streaming, and batch calls, thus offering great flexibility according to different application needs.
Built-In Utilities: Generate includes several useful features like chainlit UI, request validation, parameter checking, rate control, disk cache, and tools for agent and model calls.
Lightweight and High-Quality Code: It minimizes dependencies, incorporates native request and authentication logic for diverse platforms, and includes strict code quality measures such as type hints and comprehensive testing.

Getting Started

Installation

To start using Generate, you simply install it via pip:

pip install generate-core

Viewing Model List

Once installed, you can easily browse through the available models:

from generate.chat_completion import ChatModelRegistry

print('\n'.join([model_cls.__name__ for model_cls, _ in ChatModelRegistry.values()]))

Model API Configuration

Configuring a generative model is straightforward. For instance, configuring the WenxinChat model requires minimal setup:

from generate import WenxinChat

# Instructions on setup
print(WenxinChat.how_to_settings())

Usage

Text Generation

Generate supports basic text generation from platforms like OpenAI:

from generate import OpenAIChat

model = OpenAIChat()
output = model.generate('Hello, GPT!', temperature=0, seed=2023)

Multimodal Text Generation

It can also handle complex multimodal text tasks, like linking images with textual descriptions:

from generate import OpenAIChat

model = OpenAIChat(model='gpt-4-vision-preview')
user_message = {
    'role': 'user',
    'content': [
        {'text': 'Where is this picture from?'},
        {'image_url': {'url': 'https://example.com/image.jpg'}},
    ],
}
output = model.generate(user_message, max_tokens=1000)

Advanced Features

Structured Data Generation

from generate import OpenAIChat
from pydantic import BaseModel

class Country(BaseModel):
    name: str
    capital: str

model = OpenAIChat().structure(output_structure_type=Country)

Rate Limiting

Generate allows for controlling the request rate:

from generate import OpenAIChat

limited_model = OpenAIChat().limit(max_generates_per_time_window=2, num_seconds_in_time_window=10)

Applications

Generate supports various forms of data generation, including text, images, and audio, thereby playing a crucial role in creative and content-driven applications. Its asynchronous and streaming capabilities make it suitable for real-time applications.

UI and Execution

Generate includes a GUI for more intuitive management of tasks and models:

python -m generate.ui

In conclusion, Generate is a robust and versatile tool that simplifies the use of multiple generative platforms through seamless integration and a unified API, thereby streamlining workflows and enhancing productivity for AI-driven projects.