---
title: "SimPO - Simple Preference Optimization - Air Market"
description: "Teaching a model which answers people prefer usually means keeping a second reference model alongside it. This method drops that requirement, which makes the setup simpler and the training more effic…"
canonical: "https://airmarket.ai/app/simpo-training"
---

[Air Market](/) / [Data & Analytics](/category/data) / SimPO - Simple Preference Optimization

Skill by [Orchestra Research](/@orchestra-research)

# SimPO - Simple Preference Optimization

Align a model on preferences without keeping a second one around

## What this does

Teaching a model which answers people prefer usually means keeping a second reference model alongside it. This method drops that requirement, which makes the setup simpler and the training more efficient than DPO. Reach for it when you are doing preference alignment and want fewer moving parts.

### Top 3 things it does for you

- Align a model to preference data
- Train without loading a reference model
- Choose an alternative when DPO feels heavy

## Information

**SimPO - Simple Preference Optimization at a glance**

|  |  |
| --- | --- |
| Kind | Skill |
| Publisher | [Orchestra Research](/@orchestra-research) |
| Category | [Data & Analytics](/category/data) |
| Works with | Agent Skills - an open standard supported by 40+ agents |
| Version | 1.0.0 |
| Size | 32 KB |
| Licence | MIT |
| Repository | [orchestra-research/ai-research-skills](https://github.com/orchestra-research/ai-research-skills/tree/28f2d29236f2bade2eb92cadb2585189589a5828/06-post-training/simpo) |
| GitHub stars | 11,987 (as of Nov 2026) |

## Security report

**Verified.** AIR scanned this exact version of SimPO - Simple Preference Optimization and flagged nothing. AIR's verdict: **safe**.

**What SimPO - Simple Preference Optimization's check was run against**

|  |  |
| --- | --- |
| Commit | 28f2d29236f2bade2eb92cadb2585189589a5828 |
| Content hash | 38d0d79a0a9c0db86c8bceacfc7d3268b72189fce6faeca55401990cb12d7b5b |
| Files read | [Browse the files at this exact commit](https://github.com/orchestra-research/ai-research-skills/blob/28f2d29236f2bade2eb92cadb2585189589a5828/06-post-training/simpo/SKILL.md) |

A scan is named by the exact content it read, not by a version number. A different version is a different artifact, and nothing here applies to a version AIR has not scanned.

[Read the full security report for SimPO - Simple Preference Optimization](/security/simpo-training)

## About SimPO - Simple Preference Optimization

Simple Preference Optimization is a reference-free approach to LLM alignment, offered as an alternative to DPO with better reported performance on AlpacaEval.

Because no reference model is needed, training is more efficient than DPO. Use it for preference alignment when you want a simpler and faster path than DPO or PPO.

## What's inside

- SKILL.md - instructions
- references/datasets.md - reference
- references/hyperparameters.md - reference
- references/loss-functions.md - reference

## Questions

### Is SimPO - Simple Preference Optimization safe to install?

AIR scanned this exact version of SimPO - Simple Preference Optimization and flagged nothing. AIR's verdict: safe. The verdict is named by the content that was read, at commit 28f2d29236f2, so it cannot drift onto a version nobody scanned. The full report is at https://airmarket.ai/security/simpo-training.

### How do I install SimPO - Simple Preference Optimization?

Open https://airmarket.ai/app/simpo-training on Air Market and use the install action: Air Market hands Claude a prompt naming this exact skill, and Claude sets it up. Source, pinned to the exact commit this listing describes: https://github.com/orchestra-research/ai-research-skills/tree/28f2d29236f2bade2eb92cadb2585189589a5828/06-post-training/simpo. Install from that link, not the branch.

## Related on Air Market

- [More Data & Analytics listings](/category/data)
- [Everything by Orchestra Research](/@orchestra-research)

---

Canonical URL: <https://airmarket.ai/app/simpo-training>