Hugging Face released TRL v1.0, a post-training library implementing over 75 methods, now downloaded 3 million times monthly. Authors Quentin Gallouédec, Steven Liu, Pedro Cuenca, and Sergio Paniego describe a "chaos-adaptive design" that separates stable and experimental APIs within the same package, accommodating the rapidly evolving post-training field. The release reflects TRL's transition from research codebase to production infrastructure, with downstream projects like Unsloth and Axolotl relying on its stability.
No score is assigned. Sources and their independence are shown in the citation chain below.