Hugging Face Transformers adds Decision Transformer for offline RL with Gym checkpoints

AI Impact Summary

Decision Transformer integration on Hugging Face enables offline RL workflows to be accessed through the Transformers library and Hub, exposing pretrained checkpoints for continuous control tasks. It formalizes RL as sequence modeling conditioned on return-to-go, past states, and actions, with ready-made examples using Gym Hopper, Walker2D, and Halfcheetah environments. This lowers the barrier for researchers to experiment with return-conditioned policies and accelerates benchmarking, though production deployments will require rigorous offline data validation, evaluation, and monitoring of data-supported assumptions.