How-To Guides#
Task-focused guides for nemotron steps run byob/mcq with the mcq family.
Start with Getting Started with Building MCQ Benchmarks if you have not produced benchmark.parquet yet.
Setup and configuration#
Prepare your data
Lay out input_dir, text or Parquet inputs, and target_source_mapping.
Domain corpus files
Create per-target directories of .txt files and match them to YAML.
Model endpoints
Configure OpenAI-compatible providers for generation, judgement, expansion, validity, and filters.
Advanced workflows#
Prompt tuning
Point prompt_config at a YAML file that defines stage templates.
Skip stages
Resume with skip_until and cached Parquet files.
Workflow overview#
flowchart LR
A[Prepare data layout] --> B[Edit YAML]
B --> C[uv run nemotron steps run byob/mcq]
C --> D{Need translation?}
D -->|yes| E[translate config + passthrough]
D -->|no| F[Done]
E --> F
C -.->|iterate| G[skip_until]
G --> C