Glossary
What is TRL? Definition and meaning
Provides trainers for SFT, reward modelling, PPO, DPO and related alignment methods.
Definition
TRL: Provides trainers for SFT, reward modelling, PPO, DPO and related alignment methods.
In short: Hugging Face library for training with reinforcement learning.
When it is used
Running preference-tuning and RLHF-style pipelines.
Learn more
The full guide to TRL explains how it works, when to use it and how it compares with the alternatives, with examples. The glossary lists every term on the site from A to Z.