Hailo AI Software Suite
Hailo software products comprise a suite of frameworks and tools designed to facilitate the compilation, execution, and evaluation of neural networks on Hailo devices.
In the Model Build Environment
Dataflow Compiler
The Data Flow Compiler API is designed to compile user-trained deep learning models into Hailo binaries, which can be deployed on Hailo devices. This toolchain accepts data from various sources, including TensorFlow checkpoints, TensorFlow frozen graph files, TFLite files, and ONNX files, ultimately generating a Hailo Executable Format (HEF) binary file.
The construction process consists of several key steps: translating the original model into a Hailo-compatible format, optimizing model parameters, and compiling the model into the final binary file.
HailoRT
Hailo's software products include frameworks and tools designed for compiling, running, and evaluating neural networks on Hailo devices. The HailoRT runtime environment package comprises several essential components: the HailoRT library for inference from C/C++ programs, the HailoRT Python package for loading models and managing data, and the HailoRT CLI for device control and inference management. It also features a PCIe driver for interfacing with the device and managing firmware, a Yocto layer for integrating Hailo software into Yocto environments, Hailo firmware for device control, and the Hailo GStreamer Plugin for facilitating parallel network inference within GStreamer pipelines. Additionally, the package includes installation scripts and configuration files to streamline setup and integration.
In the Runtime Environment
Model Zoo
The Hailo Model Zoo provides a wide range of pre-trained models tailored for high-performance deep learning applications, enabling users to assess full precision accuracy, evaluate quantized accuracy using the Hailo emulator, and test performance on the Hailo-8 device. It streamlines the development process by generating Hailo Executable Format (HEF) binary files, which help create high-quality applications optimized for the Hailo-8 platform. Additionally, the Model Zoo offers retraining instructions for customizing models to fit specific datasets and includes internally developed models designed for particular use cases.
TAPPAS
Hailo TAPPAS is a comprehensive collection of application examples that showcase pipeline elements and pre-trained AI tasks. It demonstrates Hailo's system integration scenarios for specific use cases across predefined software and hardware platforms. This resource can be utilized for evaluations, reference code, and demonstrations. Hailo TAPPAS offers several advantages, including accelerated time to market by reducing development time and deployment effort, simplified integration with the Hailo runtime software stack, and a foundational starting point for customers to refine and enhance their applications.