Determined

What is Determined?

Determined is a deep learning training platform that works with PyTorch and TensorFlow. It handles distributed training, hyperparameter tuning, cluster resource management and experiment tracking, so machine learning teams can spend their time on models instead of infrastructure. You connect your existing model code through the Determined API, launch experiments from the command line or web interface, and the platform schedules them on your GPUs, with support for preemptible cloud instances to reduce spend. Besides experiments, it offers free-form commands, shells over SSH, Jupyter notebooks and TensorBoard instances on the same cluster. It can be installed on a single machine, on a managed agent cluster on premises or in AWS and GCP, on Kubernetes, or on Slurm. An open source edition and an Enterprise Edition exist; single sign-on, role-based access control and Slurm support belong to the Enterprise Edition. No price list is published on the documentation pages, so check with the vendor for Enterprise terms.

Use cases

• Train deep learning models across many GPUs
• Tune hyperparameters automatically
• Track experiments and reproduce results
• Share a GPU cluster between research teams
• Run Jupyter notebooks and TensorBoard on a cluster
• Schedule training jobs on Kubernetes or Slurm

Who it's for

• Machine learning engineers
• Data scientists
• ML platform teams
• Research labs

Why use Determined

• Keeps your existing PyTorch, TensorFlow or Keras code and adds distributed training through the Determined API.
• Pools GPUs into one shared cluster so teams queue and prioritize jobs instead of reserving machines by hand.
• Records code versions, metrics, checkpoints and hyperparameters for each experiment, which makes results easier to reproduce.
• Runs wherever your compute lives: a laptop, an on-prem cluster, AWS, GCP, Kubernetes or Slurm.

Determined features

• Distributed training without changing model code
• Hyperparameter tuning to find strong models
• Experiment tracking of code versions, metrics, checkpoints and hyperparameters
• Resource management with preemptible instances and smart scheduling
• Command-line interface for experiments, queues, notebooks and TensorBoard
• Commands and shells for free-form tasks and SSH access to cluster GPUs
• Configuration templates to share settings across workloads
• Deployment on Kubernetes, on-prem agents, AWS and GCP
• OAuth, OpenID Connect, SAML, SCIM and RBAC in the Enterprise Edition

Determined integrations

• PyTorch
• TensorFlow
• Keras
• TensorBoard
• Kubernetes
• Slurm
• AWS
• Google Cloud
• Azure Kubernetes Service
• Okta

How to get started with Determined

1. Install Determined locally, on an agent cluster, on Kubernetes or on Slurm by following the deployment guide for your environment.
2. Connect your existing model code to the Determined API and describe the experiment in a YAML configuration file.
3. Launch the experiment from the CLI; the scheduler assigns GPUs and runs the trials, using distributed training where configured.
4. Review metrics, checkpoints and hyperparameter search results, and open notebooks or TensorBoard on the same cluster.

Determined limitations

• Slurm and PBS deployment is only available in the Enterprise Edition
• OAuth, SAML, SCIM and RBAC apply to the Enterprise Edition only
• You operate the cluster yourself, whether on premises or in a cloud account

Determined pricing

Determined is open source, so you can use and self-host it for free.

Prices can change and may vary by country or exclude tax. Check the official pricing page.

Determined alternatives

The main Determined alternatives listed on Debuly are Kubeflow, MLflow, Weights & Biases, Crusoe Cloud, DVC and Modal. The table compares pricing and platforms; order is A–Z.

1. 

   Crusoe Cloud

   AI cloud with on-demand GPU instances, managed Kubernetes and Slurm, and managed inference and fine-tuning for open models.

   Developer platformsCoding
   PaidVisit Crusoe Cloud (official site)
2. 

   DVC

   Open-source, Git-based data version control that tracks datasets, models and experiments next to your code in ML projects.

   Developer platformsCodingResearch and search
   Open sourceVisit DVC (official site)
3. 

   Kubeflow

   Open source, Kubernetes-native AI platform made of modular projects for training, tuning, pipelines, notebooks and model management.

   Developer platformsAutomation and agents
   Open sourceVisit Kubeflow (official site)
4. 

   MLflow

   Open-source AI engineering platform for tracing, evaluating and deploying LLM agents, apps and classic ML models.

   Developer platformsAutomation and agentsCoding
   Open sourceVisit MLflow (official site)
5. 

   Modal

   Serverless cloud platform for running AI inference, training, batch jobs and agent sandboxes, billed per second of compute.

   Developer platformsAutomation and agentsCoding
   FreemiumVisit Modal (official site)
6. 

   Weights & Biases

   Platform for tracking ML experiments and tracing, evaluating and monitoring LLM apps and agents with W&B Weave.

   Developer platformsCodingAutomation and agents
   FreemiumVisit Weights & Biases (official site)

Determined FAQ

Sources & verification

Facts on this page come from the official sources below, each with the date we read it. Written by Debuly; the vendor did not review this page.

Image credits

• Logo: none collected yet, so a monogram is shown.
• Cover: none collected yet, so a generated tile is shown.

Spot an error? Report an issue with this listing. We also remove a logo or image on the owner's request.