ES at Scale

Evolution Strategies at Scale: LLM Fine-Tuning Beyond Reinforcement Learning

Xin Qiu, Yulu Gan et al. · ICML 2026

Read paper · arXiv:2509.24372

Reward-weighted parameter perturbations.

What you will implement

Reward standardization and ES updates.

• 104. ES Reward-Weighted Update

  Reward-weighted parameter perturbations.

  Reward standardization and ES updates.