Clean training data with GPU-powered filters.
Prepares cleaner training datasets by filtering content, removing duplicates, and redacting personal information with GPU acceleration.
When to use it
Use when preparing training data from web scrapes, deduplicating large datasets, filtering unwanted content, or curating multiple media types.
Give it training data files; it saves a curated dataset after filtering, deduplication, and personal-information redaction.
What you provide
Requires the NeMo Curator package; the GPU text-curation setup installs its text_cuda12 extra.
Lists cudf as a dependency.
Lists dask as a dependency.
This skill
nemo-curator
Installs NeMo Curator