We, at Iterait, love open-source. We build our solutions on open software and we try to return the favour by improving these project for their communities.Learn more
Naturally, we continue to develop new open software every day. Here is a growing list of our own projects available for everyone!
Emloop is a lightweight framework for machine learning with focus on modularization, re-usability and rapid experimenting.
If you need to experiment with various machine learning models and configurations yet you want to keep everything manageable and repeatable emloop may be a perfect fit.
HiPipe is a C++ library for efficient data processing. Its main purpose is to simplify and accelerate data preparation for deep learning models, but it is generic enough to be used in many other areas.
HiPipe lets you build intuitive data streams that transform, combine and filter the data that pass through. It maximizes the utilization of the provided hardware and provides an ultimate performance.
Apistrap is a tool for simple and fast REST API definitions in Flask or AioHTTP. Simply decorate a function with expected payloads and responses and let apistrap manage the payload data parsing and integrity. You can focus just on the actual functionality as all the API overhead is managed by apistrap. Finally, created APIs may be conveniently exported in OpenAPI format for client generation etc.
Serving multiple (deep learning) models with limited resource (e.g. a single GPU card) can be quite challenging. Shepherd is a smart service scheduler which enables you to run as many computationally intensive services as required.
As long as your models are emloop compliant, a simple configuration file is all you need to serve your models. Ultimately, Shepherd provides a simple unified REST API.
Although our projects may be utilized separately, their true potential is unleashed only when they are used together. From data collection, streaming, model training to management and model serving, our tools and frameworks handle it all. Everything is interconnected yet modularized and well tested in all of our projects.