skills.rest
NVIDIA
NVIDIAOFFICIAL
Explore all skills by this author →
WebsiteGitHub

recipe-recommender

Index and suggest Megatron Bridge training recipes by model, GPU count, and goal.

2.8k|332|Updated Feb 25, 2026
Search Tags:#config#recipes#training#parallelism#recipe#megatron-bridge
One-click install
npx skills add https://github.com/NVIDIA/skills --skill recipe-recommender
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: recipe-recommender
Source: https://github.com/NVIDIA/skills/tree/main/skills/Megatron-Bridge/recipe-recommender
Command: npx skills add https://github.com/NVIDIA/skills --skill recipe-recommender

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Indexes shipped recipes and helps users pick the right starting config, adjust parallelism, and avoid common pitfalls.

Core Features & Use Cases

  • Recipe discovery across library and performance recipes for Megatron Bridge.
  • Start-config recommendations based on model name/size, GPU count, and training goal (pretrain, SFT, or PEFT) with guidance on parallelism and batch sizing.
  • Pitfall avoidance and optimization tips to improve training throughput and stability.
  • Use case: a user with an 8B model on 8 GPUs wanting a quick starting point for pretraining.

Quick Start

Ask for a recommended recipe for your model and hardware, then apply the suggested starting config.

Frequently Asked Questions about recipe-recommender

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

FAQPage Schema
How do I find the right starting config for Megatron Bridge training?▼

Find a Megatron Bridge starting config by indexing shipped recipes based on your model name, GPU count, and training goal, which provides tailored recommendations across pretrain, SFT, or PEFT libraries.

Can I get recipe recommendations for pretraining an 8B model on 8 GPUs?▼

Yes, recipe recommendations for an 8B model on 8 GPUs are generated by specifying your hardware constraints and pretraining goal to retrieve a matching config with parallelism and batch size guidance.

What is the difference between library recipes and performance recipes in Megatron Bridge?▼

Megatron Bridge library recipes provide starting configs for pretrain, SFT, or PEFT training, while performance recipes focus on benchmarking and optimizing throughput using scripts under the performance directory.

How do I adjust parallelism and batch size when applying a training recipe?▼

Adjust parallelism and batch size by applying the suggested starting config from your recommended recipe, which includes tailored guidance to help you tune parameters and avoid common training pitfalls.

Does Megatron Bridge support starting configs for SFT and PEFT workflows?▼

Yes, Megatron Bridge supports starting configs for SFT and PEFT workflows by indexing library recipes that match your specific model, GPU count, and training goal across these training types.

What are common pitfalls when configuring parallelism for Megatron Bridge training?▼

Common pitfalls when configuring parallelism include mismatched batch sizes and inefficient GPU utilization, which you can avoid by using indexed recipe recommendations that provide tailored optimization tips for stability.

Related Skills

Explore More Software Engineering Skills→
yo-stevenyo-steven

recipe-recommender

Recommends Megatron Bridge training recipes based on GPU topology and goals.

Community
Intermediate
moose-codemoose-code
2

indexing-performance

Tunes HyperSync, batch sizes, and RPC parameters to optimize indexer throughput and stability for historical/live indexing workflows.

Community
Intermediate
yo-stevenyo-steven

perf-tp-dp-comm-overlap

Configure Megatron-Bridge communication overlap for tensor, data, and pipeline parallelism.

Community
Advanced
ovachieverovachiever
52

training-llms-megatron

Train large language models with Megatron-Core using configurable 3D and 4D parallelism.

Community
Advanced