Skip to main content
Deployment is where you move from experimentation to a stable production serving path. Use it when a workload deserves dedicated runtime capacity, deployment-specific controls, or a promoted trained model.

What deployment gives you

  • dedicated deployment creation and configuration
  • deployment-specific API examples and model identifiers
  • instance and settings surfaces for operating production capacity
  • a clean destination for models that come out of fine-tuning

When to choose Deploy

Choose Deploy when:
  • the workload is important enough to justify dedicated capacity
  • you need stronger control over the serving path than the shared serverless API provides
  • you want to promote a trained model into production
  • you want a stable endpoint and deployment-specific operating surfaces

Serverless vs dedicated

PathBest for
Serverless APIFast experimentation, variable traffic, broad access to hosted models
Dedicated deploymentStable production throughput, custom deployment settings, trained model promotion
  1. Start by capturing traffic from your application.
  2. Use evals to define success.
  3. Use fine-tuning if you need a task-specific model improvement.
  4. Move into deployment once the model and workload deserve their own production runtime.
  5. Keep observing the deployment in production.

Next steps

Deploy a Trained Model

Promote a completed training run into a production serving path.

Deploy a HF Model

Bring a Hugging Face model onto Inference.net.

Fine-tuning

Prepare a model that is worth promoting.

Capture Traffic

Keep production traffic visible after rollout.

Talk to an engineer

Meet with us if you want help planning deployment topology, scaling, or rollout strategy.