Principle:Marker Inc Korea AutoRAG Pipeline Runner Initialization
| Knowledge Sources | |
|---|---|
| Domains | Deployment, Pipeline_Orchestration |
| Last Updated | 2026-02-08 06:00 GMT |
Overview
An initialization pattern that loads an optimized RAG pipeline configuration and instantiates all module components for inference.
Description
Pipeline Runner Initialization converts a static YAML config (with one module per node) into a live inference pipeline. The BaseRunner reads the config, iterates through each node, and instantiates each module's implementation class with its stored parameters and the project directory. This creates two parallel lists: module_instances (live objects) and module_params (parameter dicts). Two initialization paths are available: from_yaml loads from an extracted best-config YAML file, and from_trial_folder automatically extracts and loads from a trial directory.
Usage
Use this principle after extracting the best config and before deploying. Choose from_yaml when you have a pre-saved config file, or from_trial_folder for one-step initialization directly from a trial.
Theoretical Basis
The initialization process:
- Parse the config to extract node_lines → nodes → modules
- For each node's single module:
- Look up the module class by module_type in the support registry
- Instantiate the class with project_dir and the module's params
- Store the instance and params for sequential inference execution