Skip to main content
A Tensorfuse deployment starts with application code, a Dockerfile, and deployment settings. The build step turns the Dockerfile into a container image and pushes that image to a private Amazon ECR registry in your AWS account. Hardware provisioning and application startup follow the build.

​
Inputs and output

  • Application code implements the endpoint or worker, such as a FastAPI application or a model server.
  • Dockerfile specifies the base image, dependencies, files, and startup command.
  • Deployment configuration specifies runtime resources, secrets, scaling, and readiness. These settings have a different role from installing dependencies in an image.
  • Container image is the packaged environment used when the application starts.
A successful image build does not establish that the model fits in GPU memory or that the service is ready to accept requests. Those checks happen when the application starts. Diagnose build errors separately from model-loading and readiness errors. See the deployment workflow for the documented sequence and deployment logs for troubleshooting.
Tensorfuse build architecture diagram