Add IRIS

[RUFFY-369]

Model description

IRIS (Imagination with auto-Regression over an Inner Speech) is a Reinforcement learning agent trained in the imagination of a world model composed of a discrete autoencoder and an autoregressive Transformer. IRIS learns behaviors by accurately simulating millions of trajectories.

The approach presented in the paper casts dynamics learning as a sequence modeling problem, where an autoencoder builds a language of image tokens and a Transformer composes that language over time.

The agent is introduced in the paper titled TRANSFORMERS ARE SAMPLE-EFFICIENT WORLD MODELS.

There is also a medium blog post to understand how the algorithm works.

Open source status

• The model implementation is available
• The model weights are available

Provide useful links for the implementation

The model is adapted from the official code repository of the paper.

The officially released weights can be found on this Github repository