Skip to content

Config Overview

The configuration system is split into two layers:

  • Shared components: reusable model and component configs such as graph constructor, SFT constructor, NER, OpenIE, EL, text embedding, document ranker, and wandb.
  • Workflow presets: task-level YAML files under gfmrag/workflow/config/gfm_rag/ and gfmrag/workflow/config/gfm_reasoner/.

Use this section in the following order:

  1. Start with the workflow family you want to run.
  2. Read the corresponding preset page to understand the top-level sections in that YAML file.
  3. Jump to the shared component pages when you want to swap graph construction, SFT construction, NER, OpenIE, EL, text embedding, ranker, or logging backends.

Workflow Families

gfmrag/workflow/config/gfm_rag/

Use this directory for the original GFM-RAG workflow family. It contains presets for:

  • indexing datasets
  • KGC training
  • retrieval and QA inference
  • SFT training
  • path visualization

See GFM-RAG Configs.

gfmrag/workflow/config/gfm_reasoner/

Use this directory for the G-reasoner workflow family. It contains presets for:

  • indexing datasets
  • KGC training
  • retrieval and QA inference
  • SFT training
  • SFT training with answer supervision
  • path visualization

See:

Shared Components

These pages document reusable config groups referenced by both workflow families: