Skip to navigation
DocumentationDesign Synthetic Data

Migrating from Standalone Library

View as Markdown

Data Designer configurations are mostly portable between the standalone library and NeMo Helix plugin.

What Usually Stays the Same

Most configuration code can stay unchanged:

  • Column definitions
  • Prompt templates and Jinja2 references
  • Constraints and profilers
  • Model aliases and inference parameters
  • Processors and generated dataset schema

What Changes Slightly

Most “locally-defined” aspects of Data Designer workloads have NeMo Helix equivalents:

| Area | Standalone library | NeMo Helix | | Inference | Define model providers in local YAML config files. | Register model providers through the Models and Inference Gateway services. | | Secrets | Reference local environment variables or provide plaintext secret values. | Create secrets using the Secrets API. | | Seed data | Reference local files or pass DataFrames. | Upload local seed data to the Files API or use HuggingFace. | | Personas | Download Nemotron Personas datasets locally. | Create Filesets for Nemotron Personas using nemo data-designer personas make-fileset. |

What Isn’t Supported

Some features of the standalone Data Designer library are not supported on NeMo Helix at all:

  • Custom column types / plugins
  • Local callable validators
  • Jinja NATIVE rendering engine
  • MCP Tools
  • Workflow chaining

Getting Help