dreamer_v2_atari_world_model(pretrained: bool = False, overrides: object = {})Construct the Atari configuration with all three objectives.
Model Size
Parameters
pretrainedbool= FalseNo weights are published for this family; passing
True raises.**overridesobject= {}Optional
DreamerV2Config field overrides.Returns
DreamerV2ForWorldModelingThe Atari trunk plus the objectives.
Notes
Reference: Hafner, Lillicrap, Norouzi, and Ba, "Mastering Atari with Discrete World Models", ICLR, 2021 (arXiv:2010.02193).
The actor's gradient is the score function here, resolved from the discrete action space — a one-hot offers nothing to differentiate through but a biased straight-through estimate.
Examples
>>> from lucid.models import dreamer_v2_atari_world_model
>>> model = dreamer_v2_atari_world_model(action_dim=18).eval()
>>> model.config.kl_weight, model.config.pcont_scale
(0.1, 5.0)