skills.rest
Lhhiep-maxcode
Lhhiep-maxcodeCOMMUNITY
Explore all skills by this author →
GitHub

nemotron-retrieval-recipes

Plan, debug, and evaluate Nemotron retrieval recipes for embedding and reranking workflows.

Updated Jul 30, 2026
Search Tags:#deployment#evaluation#retrieval#nemotron#reranking#embedding#dry-run
One-click install
npx skills add https://github.com/Lhhiep-maxcode/Nemotron --skill nemotron-retrieval-recipes-lhhiep-maxcode
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: nemotron-retrieval-recipes
Source: https://github.com/Lhhiep-maxcode/Nemotron/tree/main/skills/nemotron-retrieval-recipes
Command: npx skills add https://github.com/Lhhiep-maxcode/Nemotron --skill nemotron-retrieval-recipes-lhhiep-maxcode

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill helps you plan, debug, tune, evaluate, export, and deploy Nemotron retrieval recipes when you need better first-stage retrieval or better top-rank ordering in a Nemotron-based pipeline.

Core Features & Use Cases

  • Embedding workflow guidance: Use it to decide when to work on embedder training for low Recall@k, missing relevant documents, or weak candidate generation.
  • Reranker workflow guidance: Use it to choose reranker tuning when candidates are present but top results are poorly ordered and nDCG is low.
  • Safe execution and debugging: Use it to validate prerequisites, avoid secret leakage, compare checkpoint, ONNX, TensorRT, and NIM results, and keep long-running jobs and remote runs under control.
  • Use case: You have a Nemotron retrieval eval where Recall@100 is acceptable but nDCG@10 is poor, and this Skill helps you inspect the right stage, preserve the held-out split, and choose the correct next action.

Quick Start

Ask the assistant to analyze your Nemotron retrieval issue, identify whether embed or rerank is the right path, and give you the safest dry-run command plus the prerequisites to check before any expensive execution.

Frequently Asked Questions about nemotron-retrieval-recipes

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

FAQPage Schema
How do I decide between embedding tuning and reranking for a poor Nemotron retrieval evaluation?▼

Choose Nemotron embedding tuning when Recall@k is low and relevant documents are missing, and select reranker tuning when candidates are present but top results are poorly ordered with low nDCG.

What prerequisites should I check before running a long GPU retrieval or deployment task?▼

Before long-running GPU or deployment work, verify commands, inspect current repo checkout, configure secret-safe environments, and preserve fixed held-out evaluation splits.

How do I validate Nemotron retrieval recipes without executing expensive GPU jobs?▼

Run dry-run validations to check prerequisites and verify commands safely, ensuring your environment is configured correctly before committing to long-running remote execution.

Can I compare retrieval results across checkpoints, ONNX, TensorRT, and NIM boundaries?▼

Yes, you can compare results across checkpoint, ONNX, TensorRT, and NIM boundaries to debug discrepancies and ensure consistent retrieval performance across different execution backends.

What is the best way to debug weak candidate generation in a Nemotron retrieval pipeline?▼

Debug weak candidate generation by inspecting the embedder training stage, validating prerequisites with a dry-run, and analyzing Recall@k metrics on a fixed held-out evaluation split.

When should I avoid changing the held-out evaluation split during Nemotron retrieval tuning?▼

Always preserve the fixed held-out evaluation split during tuning and deployment to ensure consistent metrics interpretation and prevent data leakage across retrieval stages.

Related Skills

Explore More Software Engineering Skills→
rocklambrosrocklambros
2

enforcing-leakage-firewall

Detect data leakage in supervised machine-learning evaluation pipelines.

Community
Advanced
justcyljustcyl

tm

Manage remote GPU servers via tm CLI for commands, jobs, and monitoring.

Community
Advanced
HemantSudarshanHemantSudarshan

rag-implementation

Implement production RAG pipelines from document ingestion to generation.

Community
Advanced
Sir-chawakornSir-chawakorn
9

build-vector-search

Configure embedding models, ANN indices, and hybrid search pipelines for vector retrieval.

Community
Advanced