# ONNX

kokoro-onnx is a text-to-speech (TTS) project based on the Kokoro model and ONNX runtime. It supports English and plans to support French, Japanese, Korean, and Chinese. The model offers near real-time performance on macOS M1 and provides a variety of voice options, including whispering. The model is lightweight, approximately 300MB (around 80MB when quantized). This project is open-source on GitHub under the MIT license, facilitating easy integration and use for developers.

RMBG-2.0 is a background removal model developed by BRIA AI, aimed at effectively separating the foreground and background in images. The model is trained on a curated dataset including general stock images, e-commerce, gaming, and advertising content, making it suitable for commercial use and capable of driving large-scale content creation for enterprises. Its accuracy, efficiency, and versatility are comparable to leading open-source models. RMBG-2.0 is available as source code for non-commercial use.

OnnxOCR is a lightweight OCR model restructured from PaddleOCR, which frees itself from the PaddlePaddle deep learning training framework to achieve rapid inference speeds. The model supports inference in over 80 languages and, after being converted to an ONNX model, its inference speed is up to 5 times faster than when using the PaddlePaddle framework. OnnxOCR is independent of the deep learning training framework and can be directly deployed, making it suitable for scenarios with limited computational power but a need for accuracy, and it can be deployed on both ARM and x86 architectures.

transformers.js is a JavaScript library designed to offer advanced machine learning capabilities to web applications. It allows users to run pre-trained Transformers models directly in the browser without server support. The library utilizes ONNX Runtime as the backend, supporting the conversion of PyTorch, TensorFlow, or JAX models into ONNX format. transformers.js is equivalent in functionality to the Hugging Face transformers Python library, providing similar APIs to ease the migration of existing code to the web side.

Phi-3 Mini is a lightweight top-tier open-source model built upon the synthetic data and filtered websites used by Phi-2, focusing on high-quality inference-intensive data. This model belongs to the Phi-3 series and the mini version has two variants supporting 4K and 128K context lengths. The model has undergone rigorous enhancement processes, including supervised fine-tuning and direct preference optimization, to ensure precise instruction following and robust security measures. These ONNX-optimized Phi-3 Mini models run efficiently on CPUs, GPUs, and mobile devices. Microsoft has also introduced the ONNX Runtime Generate() API, simplifying the usage of Phi-3.

Phi-3 Mini is a lightweight, advanced open-source large-scale model constructed on synthetic data from Phi-2 and filtered website data, committed to providing high-quality, inference-intensive data. The model has undergone a rigorous enhancement process, combining supervised fine-tuning and direct preference optimization to ensure strict adherence to instructions and robust security measures. This repository provides an optimized ONNX version of Phi-3 Mini, which can be accelerated on CPUs and GPUs using ONNX Runtime, supporting servers, Windows, Linux, Mac, and various platforms, with the best precision configuration tailored for each platform. ONNX Runtime's DirectML support allows developers to achieve large-scale hardware acceleration on Windows devices with AMD, Intel, and NVIDIA GPUs.

sd4j is a Stable Diffusion inference Java implementation using ONNX Runtime, optimized and ported from a C# implementation. It features a graphical user interface for generating images repeatedly and supports negative text input. Its purpose is to demonstrate how to use ONNX Runtime in Java and provide best practices for achieving good performance with ONNX Runtime. We will keep it synchronized with the latest ONNX Runtime version and update it appropriately as performance-related ONNX Runtime features become available through the ONNX Runtime Java API. All code is subject to change as this is a code example, and no APIs should be considered stable.