Skip to content

📦 Model and Job Hosting on Accelerator

The Accelerator platform enables not only the execution of computational routines and jobflows but also their hosting and distribution as reusable, shareable scientific services. Whether you are building a predictive model, data harmonization workflow, or policy simulator, Accelerator allows you to publish and manage these jobs with full access control.


🎯 What Does "Hosting a Model" Mean?

In the context of Accelerator, hosting refers to:

  • Defining a routine or jobflow to specify computation
  • Providing a parameterizable interface for inputs and configuration
  • Assigning proper access policies to determine who can use it
  • Making it discoverable and reusable within and beyond your project

Think of it like publishing a reproducible simulation or processing workflow that others can run on demand, using their own inputs but your encapsulated logic.


🧱 Hosting Components

To host a model or job on Accelerator, you define and publish the following:

  1. Routine or Jobflow Entity
    Encapsulates the logic, environment, and flow of your computational workload.

  2. Input Mapping Schema
    Defines where data is expected to come from (e.g., GUI file selection, cloud path).

  3. Configuration Schema (conf)
    JSON Schema that defines the parameters users must fill before dispatch.

  4. Access Policy
    Determines whether the job is private, shared with a group, or publicly executable.

  5. (Optional) Secrets Schema
    Define sensitive credentials (API keys, tokens) users may be required to provide.


🛠️ Example Use Case: Climate Downscaling Model

Imagine you’ve built a climate model downscaling pipeline consisting of:

  • Step 1: Input data preprocessing (routine)
  • Step 2: Regional model execution (routine)
  • Step 3: Post-processing and export (routine)

You’ve wrapped this logic into a Jobflow, tested it thoroughly, and defined:

  • Input files: Provided by the user at runtime via GUI
  • Config options: scenario, target_region, base_year, etc.
  • Secrets: None (optional)
  • Output mappings: Writes to user-designated acc:// location

You can now host this jobflow as a reusable model.


🧭 Execution Experience for Other Users

Users who have permission to execute this hosted job will:

  1. See the hosted model listed in their project space or shared public catalog
  2. Select input data directly from file explorer (selected_files, selected_folders)
  3. Fill in the dynamic form generated from your JSON Schema
  4. Launch the job with one click
  5. Monitor execution, logs, and results in the GUI

✅ All jobs inherit your fixed logic and environment, while accepting variable inputs and configuration.


🔒 Access Control and Visibility

As the author, you define how the model can be used:

ScopeWho Can Access?
PrivateOnly you
Project TeamMembers of the current project
PublicAnyone with platform access
Link-BasedAnyone with the special access link

Access control applies both to viewing and executing the job.


🔄 Model Versioning

You can:

  • Update hosted job logic over time
  • Maintain multiple versions of the same model
  • Deprecate or archive older versions
  • Track usage and run history

Combined with Git + DVC-backed logic (see Running Simulations), this ensures long-term reproducibility.


🌐 Embedding Hosted Jobs in Web Portals

You can expose your hosted jobs as public services by embedding launch widgets or links in external tools. These can be used to:

  • Provide public policy simulation interfaces
  • Expose predictive models to non-technical users
  • Share reproducible analysis tools in scientific collaborations

🔗 Embedding can be done via shareable links, GUI widgets, or portal API integration.


🧬 Ideal Use Cases

ScenarioHosted Job Purpose
Climate Model RunnerAllow users to downscale climate projections
Data Harmonization PipelineStandardize incoming data from varied sources
Environmental Impact AnalyzerSimulate emissions based on input scenarios
Crop Yield Estimation ModelPredict outcomes for agricultural scenarios
Geospatial Post-Processing ToolConvert GeoTIFF to Cloud Optimized formats

🚀 Summary

Hosting jobs and models on Accelerator enables:

  • Publication of repeatable scientific computation
  • Simplified access via GUI-based forms
  • Controlled sharing with access rules
  • Execution without installing software
  • Long-term reproducibility and transparency

Accelerator makes it easy to package, publish, and share your science.