Hugging Face TRL v1.18 makes AsyncGRPOTrainer stable

Hugging Face's TRL v1.18 promotes AsyncGRPOTrainer to stable, enabling GRPO's generation and training phases to run in parallel on separate GPUs.

2026-11-13 ~ 2026-11-14 · 2 related posts

• AsyncGRPOTrainer now stable in TRL v1.18.0: generation and training on separate GPUs — SergioPaniego · 2026-11-13
• AsyncGRPOTrainer goes stable in Hugging Face TRL v1.18 — SergioPaniego · 2026-11-14