Introducing 🚀Airtrain.ai – our free batch evaluation tool for Large Language Models. Get started for free!

Retrain and ship
ML models
80% faster

The open-source orchestrator loved by ML teams

Sematic is the easiest and fastest way for ML teams to build and execute training pipelines across their dev box and cloud infrastructure.

Orchestration
Traceability
Versioning
Reproducibility
Visualizations
Metrics
The open-source

Continuous
Machine Learning Platform

End-to-end Python pipelines from your laptop to your cloud cluster in minutes. Tracked, reproducible, visualized.

Supported by
Sematic Dashboard screenshot
$ pip install sematic
Successfully intalled sematic-0.31.0
$ sematic start
Starting Sematic ...
Visit Sematic at http://127.0.0.1:5001
$ sematic run examples/mnist/pytorch
Running example examples/mnist/pytorch ...

Get started on your laptop in minutes

Get started on your local machine by simply installing Sematic, launching the dashboard, and running one of the included example pipelines.

When you're ready to scale up, learn how to deploy Sematic to your Kubernetes cluster.

Read the docs

End-to-end pipelines for Continuous Learning

An end-to-end pipeline

Implement end-to-end pipelines of arbitrary complexity to easily retrain your models when new labeled data is available.
Sequence data processing jobs with model training and evaluation to automate and accelerate your workflows.

Run on your local dev box or submit to your Kubernetes cluster seamlessly.

Python-first declarative orchestration

Use the power of Python functions to define all aspects of your pipelines.

No messy YAML templating, Jsonnet, or esoteric DSL. Just Python, for everything.

Create arbitrary complex dynamic DAGs with looping, conditional branching, nesting, and more.

  • Business logic
  • DAG
  • Configuration
  • Resource requirements
Get Started
import sematic

@sematic.func(
    resource_requirements=GPU_RESOURCE_REQS,
)
def train_model(
    dataloader: DataLoader,
    config: TrainConfig,
    device: torch.device,
) -> nn.Module:
    model = Net().to(device)
    _train_model(model, config, dataloader)
    return model

@sematic.func(resource_requirements=GPU_RESOURCE_REQS)
def evaluate_model(
    model: nn.Module, dataloader: DataLoader
) -> EvaluationResults:
    results = _evaluate_model(model, dataloader)
    return results
Using the Sematic SDK
Screenshots of the Sematic Dashboard

Track, version, visualize everything, all the time

Inputs and outputs of all steps are persisted as source of truth and visualizable in the dashboard.

Dataframes, models, configuration dataclasses, images, metrics, plots and figures. You name it, Sematic tracks it and displays it for you in the UI.

Rerun pipelines from the UI, from scratch or from any point. Cache results and implement fault tolerance for greater reliability.

Read the docs

Iterate seamlessly between your dev box and your cloud cluster

See the impact of your code change at scale within minutes.

Change code, test on your dev box on a small amount of data, then run the same code at scale in your cloud environment.

Sematic packages your local environment at runtime (code, pip dependencies, static libraries, other dependencies, etc.), ships it to your cluster and orchestrates your pipeline.

Get Started
Illustration of iterative development
Illustration of Sematic integrations

Integrates with your stack

Sematic sits in the middle of your stack and integrates with the tools you already use.

Machine Learning libraries, cloud tools, observability services, productivity tools, you name it.

Sematic's plug-in model means you can build support for new integrations any time.

Learn more

Our users love us

Join our community of forward-thinking ML Engineers, Data Scientists, Infrastructure and Platform Engineers.

"Sematic gives us unparalleled visibility into our ML pipelines (artifacts, logs, errors, source control, dependency graph, etc.) while keeping the SDK and GUI simple and intuitive.

It provides just the right level of abstraction for ML engineers to focus on business logic and leverage cloud resources without requiring infrastructure skills.

Sematic is the kind of pipelining tool used by ML teams at Uber, now available to Voxel and everyone else."

Anurag Kanungo

CTO, Co-Founder at Voxel

"Sematic is simple but powerful. In a few minutes I was able to get a template pipeline running, and tailor it to my specific needs and view everything in a surprisingly informative, straightforward UI.

The level of debugability but scalability that Sematic provides is awesome; it is barely opinionated (just add a function decorator) but hugely flexible.

I can easily see it become a standard tool for Machine Learning orchestration, pipelining, and even scaling."

Anton Bongio Karrman

ML Engineer at Ghost Automation

"What I love about Sematic is that it's just Python. It's easy to implement with existing code, but also encourages better project design and thinking about the data being passed between steps.

It saves me a couple of hours every time I run the pipeline, which allows me to experiment more, and get better results."

Blaine Bateman

Machine Learning Consultant
Author of The Supervised Learning Workshop

Subscribe to our mailing list

Receive release notes, updates, tips and news straight into your inbox.
Average frequency: every couple of weeks. No spam ever.

Subscribed, thank you!
Oops! Something went wrong while submitting the form.