Our mission is to make training artificial intelligence transparent, accessible, and cost-efficient.
Track your experiments, debug your models, and manage computation servers. It’s made for you as a single developer working completely offline and teams with real-time collaboration tools out of the box.
Cover the full lifecycle of machine learning experiments.
Better understand models by using model debugging and analytical tools
Oversee your computation infrastructure in clusters and utilise them efficiently.
High fidelity real-time user interface with collaboration tools.
Define experiments in an unified YAML format or use the pythonic way with the Deepkit Python SDK.
First interactive model debugger for Keras with Tensorflow 1, Tensorflow 2 Keras, and Pytorch.
While we have special support for Python and Tensorflow/Pytorch, you can run experiments in any language.
With the integrated cluster management you can run executing experiments on your own servers with just one click.
Split your experiment in multiple tasks and pipe outputs to decrease execution time and costs.
Run your experiment in Docker. Deepkit builds, runs, monitor, and assigns NVIDIA GPUs automatically.
Work completely offline, execute the experiment local exactly like it would in the cloud, and go online anytime.
Connect your Git with a Deepkit Team Server and run experiments directly via the app from Git source or via hooks.
Display thousands of experiments, categorise and label them, create quick filters, or run powerful search queries in the experiment list. Important metrics of filtered experiments are available in real-time right at the bottom.
Compare additional to the table list many experiments in a side-by-side view. Different values are highlighted and file diffs can be shown directly one click away.
The cockpit that enables you to see quickly how your experiment is going.
Add custom metrics via the Python SDK and display them in real-time.
Deepkit adds hardware metrics of your CPU, memory, network, disk io and GPUs automatically.
All your experiment files are automatically added. View, compare, or download.
Via the Python SDK you can upload additional files aka artifacts to make sure nothing is lost.
Further artifacts like weights in folders can be picked up automatically.
Add custom plots, images, json, text, or numpy arrays as insights to an experiment on a timeline.
Display the content of each timeline entry step-by-step to see how your model evolves.
The integrated model debugger enables you to not only see the architecture of your model any time but also to look into the model in real-time while it trains. Many layers can be visualized right next to the histogram of your output, weights, and biases.
A recorder enables you to record multiple snapshots of the model at different states.
Support for Keras with Tensorflow, Tensorflow 2 with tf.keras, and Pytorch via the Python SDK.
Integrated real-time issue tracker with kanban board enables you to quickly get an overview of work to be done and what your team is doing.
Write notes, reports, protocols, or just your findings alone or together with your colleagues in real-time like in Google Docs.
Connect any computation server to Deepkit by entering its SSH credentials. You can group computation server in multiple clusters.
Deepkit schedules your experiments according to your settings, uploads all experiment files and starts it automatically.
You can manage each node in your cluster, view current resources, utilisation, GPU stats. Docker images and containers are shown as well for you to decide which to remove to save disk space.
