Reference#

Specifications grounded in src/nemotron/steps/byob.

Outputs#

Output files

Seed, stage cache, raw and final Parquet paths.

Output Files
Troubleshooting

Common configuration errors, missing caches, filtering, and endpoint issues.

Troubleshooting

Configuration#

Generation YAML

Required keys for ByobConfig.from_yaml.

Generation Configuration Reference
Translation YAML

ByobTranslationConfig.from_yaml requirements.

Translation Configuration Reference

Source benchmarks#

Allowed Hugging Face datasets

Identifiers and default subsets from runtime/constants.py.

Supported Hugging Face Benchmarks