TRL CLI

TRL (Transformer Reinforcement Learning) CLI is Hugging Face's unified command-line interface for LLM post-training. TRL v1.0 introduced a standardized config system that makes fine-tuning accessible from the terminal. Supports all major alignment algorithms — SFT (supervised fine-tuning), DPO (direct preference optimization), GRPO, KTO, and reward modeling. Designed for ML engineers who want reproducible training pipelines without writing boilerplate Python scripts.