Valohai Logo


Machine Orchestration, Version Control and Pipeline Management for Deep Learning

From feature extraction
to training and deployment

Valohai infrastructure overview

Scale fast and without effort

Valohai supports massive-scale concurrency on top of AWS, Microsoft Azure, Google Cloud Platform & on-premises hardware (e.g. OpenStack). Just click a button and launch your code within Dockers containers on your hardware of choice.

Automate your version control

Fulfill regulatory compliance without added work. Valohai automatically tracks all your experiments, with a clear picture of how each model was trained, from data to parameters & statistics to algorithm. Rerun previous experiments anytime.

Pipeline management

Don’t worry about environments, configurations or shutting down servers when your training is done. Streamlined and expandable Valohai API allows you to concentrate on trials & mastering your models!

Interested in the latest updates to Valohai?
Check out the latest patch notes (December 3, 2019) or browse all patch notes.

Track and visualize
every experiment

From Jupyter to Scripts

Data Scientists should be able to use the tools and frameworks they like and the ML platform should integrate with your way of working. This is at the core of Valohai!

Valohai lets you work in Jupyter notebooks, with scripts on your local machine, or on a shared project in a Git repository. In any programming language, with any framework – the choice is yours. Valohai takes care of machine orchestration and automatic version control so you can concentrate on your code and your data.


We believe that effective version control is the only way to achieve reproducibility, regulatory compliance, an audit trail & quick results.

Whether from today, or 10 years from now, you’ll be able to select a deployed model and clearly trace back through its hyperparameters, training data, script version, associated cost, sibling models & even the team members involved in training it.


You’ll see everything in real-time as your trainings progress, no longer stuck manually launching models and keeping track of CSV files. Get visual feedback on everything from a single model’s performance to a convergence of several parallel hyperparameter sweeps. See how your parameter sweeps are progressing while comparing competing models by accuracy, depth, or any custom parameter. You can also output custom parameters into stdout and see it graphically in the Valohai web interface.


Valohai works with any runtime you have and runs any machine learning code you write. Unlike other deep learning tools, we don’t tie you down to one vendor (not even to ourselves – even the configuration format is open source).

alt text Run your TensorFlow, Keras, CNTK, Caffe, Darknet, DL4J, PyTorch, MXNet, or anything from bash scripts to C-code in your Docker wrapper of choice. Store your training data and labels in an Azure Blob, AWS S3 bucket, or your own FTP server. Access your code in any public or private Git repository and run it on your cloud or on-premises hardware of choice.

Read more about Valohai open standards ➜


Valohai puts the same tools and industry-leading best practises at your fingertips used by powerhouses like Uber, Netflix, AirBnB and Facebook for managing their internal machine learning pipelines.

Valohai’s streamlined machine learning pipeline ensures that steps integrate together, regardless of who wrote the code or which language or framework was used. Generate images with Unity, transform in custom C-code, train with TensorFlow in Python, Deploy to a Kubernetes cluster. Everything works out of the box!

Valohai API


Everything in Valohai is built API-first for easy integration of your ML pipeline into your existing software pipeline, e.g. through Jenkins or any other continuous integration platform.

Powerful machine orchestration


Valohai lets you scale up vertically and horizontally to do distributed learning and parallel hyperparameter sweeps at the speed of light (in an ethernet cable). Run your model in parallel on a hundred GPUs or tell Valohai to sweep through different hyperparameters to find the best model for your data in parallel on dozens of TPUs. Valohai is built for finding and optimizing your model for big data and immense models that scale with you, as you grow from data exploration to production.

Infinite scale!

Zero-Setup Infrastructure

Train your models in the cloud or on your own server-farm with the click of a button, the call of an API, or a CLI one-liner. Valohai enables you to use the right amount of processing units - maximizing your results while saving time & money.


Give your data science team full transparency into how models have been trained throughout their history right up until what every team members is working on today.

Share projects

Assign team members to projects, share data science code, tag each other in free-form research notes and collaborate more efficiently.

Share results

Automatically share all experiments in a project with the rest of the team. View models in real-time as they converge and spin off new trainings.

Share models

Trace back from a deployed model to its training data, hyperparameters, code, environment (software & hardware), and more.


Protect your data

Protect your business and your customers' data by hosting all of your assets in your own cloud environment or on-premise data storage. Valohai supports any web storage from Amazon S3, Azure Blob Storage, an HTTP endpoint or a directory in your intranet.