agentsclimarketplace
Sign in

Run2 simpo implementation

Skill cxcscmu/SkillLearnBench/skills/b2-self-feedback-gemini-3.1-flash-lite-preview/nlp-paper-reproduction/run2_simpo-implementation

Detailed SimPO implementation logic covering reward formulation and margin-based loss computation.From its SKILL.md

Install
npx -y skills add cxcscmu/SkillLearnBench --skill run2_simpo-implementation

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

0.5 KB, 105 tokens by cl100k_base, as published. Nobody here has run it

SimPO Loss Definition

SimPO reward formulation:

  1. chosen_rewards = beta * log_prob_chosen
  2. rejected_rewards = beta * log_prob_rejected

Loss for sigmoid-based SimPO (Equation 6): loss = -log_sigmoid((chosen_rewards - rejected_rewards - gamma) / beta)

  • Ensure beta is set to control the reward scale.
  • The gamma_beta_ratio corresponds to gamma/beta in the paper implementation.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.