SimPO Loss

SkillAI & models

SimPO (Simple Preference Optimization) loss computation for LLM alignment without a reference model.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the SimPO Loss skill

What this skill tells your AI

The instructions your AI receives, as published by cxcscmu/skilllearnbench in skills/b1-one-shot-claude-opus-4-6/nlp-paper-reproduction/simpo-loss/SKILL.md and read by ahel’s review.

Overview

SimPO is a reference-free preference optimization algorithm. Its key innovation is using the average log probability of a sequence as the implicit reward, plus a target reward margin γ.

Loss Formula (Eq. 6 from the paper)

L_SimPO = -E log σ(β/|yw| · log πθ(yw|x) - β/|yl| · log πθ(yl|x) - γ)

Since the log probabilities passed to simpo_loss are already length-normalized (average log prob), the loss simplifies to:

logits = β * policy_chosen_logps - β * policy_rejected_logps - γ

where γ = gamma_beta_ratio * beta.

Loss Types

  • sigmoid (default): losses = -log σ(logits) * (1 - label_smoothing) - log σ(-logits) * label_smoothing
  • hinge: losses = relu(1 - logits)

Rewards

  • chosen_rewards = β * policy_chosen_logps
  • rejected_rewards = β * policy_rejected_logps

Default Hyperparameters

  • β = 2.0
  • gamma_beta_ratio = 0.25 (so γ = 0.5)
  • label_smoothing = 0.0
  • loss_type = "sigmoid"

Signals

GitHub stars
83
Forks
5
Last commit
Jul 2026
Advanced
Catalog kind
skill
Gateway key
simpo-loss
Source
github.com/cxcscmu/skilllearnbench