skills.rest
NVIDIA-NeMo
NVIDIA-NeMoOFFICIAL
Explore all skills by this author →
WebsiteGitHub

perf-moe-optimization-workflow

Optimize MoE model training throughput by addressing memory, communication, and compute bottlenecks.

852|445|Updated May 21, 2025
Search Tags:#performance#scalability#optimization#training#deep-learning#moe
One-click install
npx skills add https://github.com/NVIDIA-NeMo/Megatron-Bridge --skill perf-moe-optimization-workflow
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: perf-moe-optimization-workflow
Source: https://github.com/NVIDIA-NeMo/Megatron-Bridge/tree/main/skills/perf-moe-optimization-workflow
Command: npx skills add https://github.com/NVIDIA-NeMo/Megatron-Bridge --skill perf-moe-optimization-workflow

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill provides a systematic workflow for optimizing the training performance of Mixture-of-Experts models, helping users enhance throughput and reduce bottlenecks.

Core Features & Use Cases

  • Performance Tuning: Guides users through memory, communication, and compute optimizations for MoE models.
  • Workflow Phases: Offers a clear, phased approach from fitting models into memory to profiling and scaling.
  • Use Case: Ideal for deep learning engineers tuning large-scale MoE training jobs on GPU clusters, such as Megatron-LM, to improve efficiency and scalability.

Quick Start

Describe the workflow steps for diagnosing and enhancing MoE model training performance.

Frequently Asked Questions about perf-moe-optimization-workflow

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

FAQPage Schema
How do I optimize MoE training throughput for large-scale deep learning models?▼

To optimize MoE training throughput, systematically address memory, communication, and compute bottlenecks by tuning, profiling, and scaling your distributed GPU workflow across phased optimization steps.

What is the best way to resolve memory and communication bottlenecks in Mixture-of-Experts training?▼

Resolving MoE bottlenecks involves a phased approach that fits models into memory, profiles distributed GPU performance, and scales Mixture-of-Experts workflows to maximize training efficiency.

Can I use this workflow to tune Megatron-LM training jobs on GPU clusters?▼

Yes, this workflow is ideal for deep learning engineers tuning large-scale Megatron-LM MoE training jobs on distributed GPU clusters to improve efficiency and scalability.

Why does MoE model training performance plateau during distributed GPU scaling?▼

MoE training performance plateaus due to memory, communication, and compute bottlenecks that require systematic profiling and tuning to overcome during large-scale distributed GPU scaling.

What are the main phases for diagnosing Mixture-of-Experts performance issues?▼

The main phases for diagnosing MoE performance involve fitting models into memory, profiling training workflows, and scaling distributed GPU environments to isolate and resolve throughput bottlenecks.

Related Skills

Explore More Software Engineering Skills→
JuanMS20JuanMS20

huggingface-accelerate

Configure distributed PyTorch training across multi-GPU and multi-node devices.

Community
Advanced
ccfccf

vector-index-tuning

Tune HNSW parameters and quantization for large-scale vector search indexes.

Community
Advanced
NVIDIANVIDIA
2.8k

perf-analysis

Coordinate profiling delegation and generate structured performance analysis reports.

Official
Advanced
ricablericable

training-llms-megatron

Train large language models with distributed parallelism across multi-GPU clusters.

Community
Advanced