Hugging Face has unveiled TRL, a new full-stack library developed as an extension of its transformers collection. It focuses on training transformer language models and stable diffusion models with reinforcement learning.

The product allows users to easily fine-tune language models using techniques like supervised fine-tuning (SFT), reward modeling (RM), and proximal policy optimization (PPO). These methods optimize the models based on a specified reward signal determined by human experts or other models. 

Hugging Face, a French-American firm, provides access to a vast library of transformer-based foundation models developed by Google, OpenAI, and Meta. The models are accessible through its Transformers library, facilitating model loading and fine-tuning pre-trained models on custom datasets.