nemotron-customize

Plans NVIDIA Nemotron customization pipelines and generates minimal YAML/config/run commands from step manifests.

1.8k|373|Updated Oct 3, 2025

Search Tags:#nemotron #model evaluation #data curation #yaml configuration #pipeline planning #training workflows #checkpoint conversion

One-click install

npx skills add https://github.com/NVIDIA-NeMo/Nemotron --skill nemotron-customize

Or copy as Structured Prompt for Agent▼

Please help me install this Agent Skill.
Skill: nemotron-customize
Source: https://github.com/NVIDIA-NeMo/Nemotron/tree/main/skills/nemotron-customize
Command: npx skills add https://github.com/NVIDIA-NeMo/Nemotron --skill nemotron-customize

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill turns a model-customization request into a repo-native Nemotron pipeline plan and runnable commands, avoiding manual step discovery and incorrect wiring.

Core Features & Use Cases

Pipeline planning from the NeMo/Nemotron step catalog: chooses the right steps for SFT, PEFT/LoRA, alignment (DPO/RLVR/GRPO/RLHF), data curation/translation, BYOB/MCQ benchmark preparation, checkpoint conversion, ModelOpt optimization, and evaluation.
Artifact-aware DAG and wiring validation: validates artifact compatibility (consumes/produces) so the output of one step matches the input requirements of the next.
Minimal config generation: produces only the YAML overlays/configs needed to run existing checked-in steps, not new reimplementations.
Safety and execution guardrails: blocks remote commands when required env/TOML profiles are missing and scopes shell usage to repo-safe validation patterns.

Quick Start

Use the nemotron-customize skill by invoking it from the Nemotron repo root for a specific training or conversion request, for example: ask your agent for the translate pipeline command for your input/output paths and hardware/backend constraints.

Frequently Asked Questions about nemotron-customize

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I plan a Nemotron SFT or PEFT training pipeline?▼

Planning a Nemotron alignment pipeline uses DPO, RLVR, GRPO, or RLHF steps from the repo catalog. It validates artifact compatibility between steps and generates the minimal YAML overlays required to run the workflow.

How does artifact validation work when wiring Nemotron training workflows?▼

Checkpoint conversion and ModelOpt optimization steps are included in the pipeline plan. The skill maps your intent to the correct repo-native steps and outputs the necessary YAML configs and shell commands.

Why does my Nemotron pipeline plan block execution due to missing environment profiles?▼

Pipeline execution blocks when required environment TOML profiles are missing to prevent guessing. This guardrail scopes shell usage to repo-safe validation patterns and requires strict repo-root-relative path resolution.

Can I use this to generate configs for BYOB and MCQ benchmark preparation?▼

Yes, you can generate configs for BYOB and MCQ benchmark preparation and translation. The skill maps your data curation request to existing checked-in steps and produces minimal runnable configurations.