Library / SDK
Fine-tuning & Training
TRL
1Train 2language 3models 4with 5reinforcement 6learning.
the six Ws · specification
W1
Who
Maintained by Hugging Face.
W2
What
Provides SFT, DPO, PPO, and GRPO trainers for language models.
W3
Where
A Python library running on your GPUs.
W4
When
When post-training or aligning an open model.
W5
Why
The standard toolkit for RLHF and preference tuning.
W6
With
PyTorch, Transformers, PEFT, and Accelerate.
W7
Watch
Official Hugging Face library, Apache-2.0, no credentials; trains on your own hardware.
alternatives
works with