Version control
Version control in data science is more than just code and binary data.
In order to reproduce your experiments you also need to reproduce your
hyperparameters, library versions and hardware. Valohai automatically
stores all of this.
Version Control
Automatically version everything you run. Version control for machine learning is non-trivial compared to software engineering and a must for real-life ML.
Audit trail
Valohai tracks every experiment you and your team conduct,
and you can see the lineage of experiments, data sets, code,
parameters, and more resulted in your final model.
Experiment comparison
Pick any hyperparameter sweep or any amount of randomly selected
experiments and compare them in a graph or table format. It's
easier than you think!
Data dictionary
Valohai automatically creates a data dictionary out of all your
input and output datasets. At the click of a button on any data
asset it'll visualize the lineage of the data from original source
to all transformations.
Read more
Model traceability
Pick a model in inference and see who the original author is, when
it was done, what data&code went into it and redo it with the same
parameters and environment at the click of a button. You'll even see
the log files from the experiment months or
years after.
Zero setup infra
MLOps is hard. Valohai makes operations transparent and easy so that you can
concentrate on models, code and data.
MLOps out-of-the-box
Setup and maintain your ML infrastructure from one place, automate
machine launch and shutdown to save costs and track every
experiment. Don’t waste time on ML grunt work and start building
models today!
Read more
Git, local code and notebooks
Code in notebooks, scripts or shared git projects in any language or
framework of your choice. Use any data storage on-prem or in the
cloud. Use the tools you already use today, with 99%
less hassle.
Hyperparameter sweeps
Visualize parallel hyperparameter tuning runs in real time, run
large distributed learning runs at the click of a button and
automate production-ready pipelines. All without breaking a sweat!
Hosted notebooks
One central hub for your Jupyter notebooks with asynchronous
hyperparameter sweeps. Start one run, make changes and start the
next. Roll back to your previous runs at any time and continue.
Exploration is trial and error but there is no reason you should
lose track of what you've done.
Jupyter notebooks & Valohai
Tool agnostic
Use the tools you love, across teams. Jupyter notebooks, PyCharm, VS
Code, Tensorflow, Keras, Darknet, DL4J, Python, Scala, R and
whatever tech your throw at it – it all works. (Because it's built
on Docker!)
Hardware agnostic
Valohai Supports AWS, GCP, Azure, Openstack and any
on-premises hardware.
API, CLI, and WebUI
Use the WebUI or the CLI for your daily work. Or expand
functionality and integrate with your own CI/CD pipeline through the
open API.
Read more
Pipeline management
After exploration and experimentation you will want to productionalize
your work through ML pipelines. Valohai lets you build complex
pipelines that you can automate.
Workflow automation
Data fetching, preprocessing, synthetic data generation,
hyperparameter sweeps and deployment for AB testing of models are
all possible cases in your ML pipeline. Valohai lets you define,
run and maintain traceability for all steps throughout your
lifetime.
Data Streaming
For datasets of 100Gb and larger, data streaming becomes an
alternative – especially if for compliance reasons you're unable to
host the data next to your computation units. Valohai lets you
stream the data, start training and continue while the data is
streaming.
Deployment
Valohai will deploy a chosen model inside
your Kubernetes cluster. No need to know how clusters work or what
Kubernetes is – as a data scientist you just pick your code and
model and Valohai will give you an HTTP endpoint you can use.
Structured and un-structured data
Valohai works with both structured and unstructured data,
from textual data to big data with 8K video streams. Or anything
in between.
Team management
Standardized workflows across your whole team ensure
transparency. Valohai lets you share projects between teams and
organizations. Your super admins can monitor costs, resource usage
and more at a project, team, person or single experiment level.
Shared projects
Machine Learning is team work, and you should be able to share
projects easily. Valohai lets you build around projects that are
accessible by you, your team or your entire organization.
Teams and groups
Valohai's access control works around the concept of teams and
organization. Define an organization, invite people and manage who
can see and do what. You can also allow only
readable or writable data source for specific organizations.
Anything is possible!
Monitoring
Monitor at a high level or check at a detailed level what your
colleagues have been working on, so you don't do the same thing
twice. Or why not just check what you did before the weekend to
quickly pick up from where you left off.
Cost management
Valohai lets you drill down to the cost on every project, team and
single experiment level. Know exactly how much money was spent and
where!
Privacy
Valohai is an ML platform built for enterprises with privacy and
security as driving principles.
SSO, 2FA and AD
Valohai supports Single Sign-On (SSO), Two Factor Authentication
(2FA) and Azure Active Directory (AD). Use your way of
authenticating with the platform!
On-premises installations
Need to run everything on-premises in an air-gapped environment on
the first lunar base? No problem! Even less exotic installations are
possible...
Private cloud
Duh... who in their right mind would want to use some public startup
cloud anyway? Of course, it'll run in your own cloud.
Private projects
Private, shared, whatever you like. GitHub, GitLab, Bitbucket
– everything works – only for you, my friend!
