> ## Documentation Index
> Fetch the complete documentation index at: https://kalarislabs.com/llms.txt
> Use this file to discover all available pages before exploring further.

# moe-training — AI agent skill for multimodal and emerging

> Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace.

# `moe-training`

> Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace. Use when training large-scale models with limited compute (5× cost reduction vs dense models), implementing sparse architectures like Mixtral 8x7B or DeepSeek-V3, or scaling model capacity without proportional compute increase. Covers MoE architectures, routing mechanisms, load balancing, expert parallelism, and inference optimization.

**Category:** [multimodal-and-emerging](/skills#multimodal-and-emerging) · **License:** MIT · **Version:** 1.0.0

## Install

```bash theme={null}
npx research-agent-skills install moe-training
npx skills add KalarisLabs/research-agent-skills --skill moe-training
```

## When to use it

Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace. Use when training large-scale models with limited compute (5× cost reduction vs dense models), implementing sparse architectures like Mixtral 8x7B or DeepSeek-V3, or scaling model capacity without proportional compute increase. Covers MoE architectures, routing mechanisms, load balancing, expert parallelism, and inference optimization.

## Full playbook

Read [SKILL.md](https://github.com/KalarisLabs/research-agent-skills/blob/main/skills/moe-training/SKILL.md) for the complete workflow, references and any scripts. The agent installer copies the full skill folder.