dreamer_v3_12m(pretrained: bool = False, overrides: object = {})Construct the smallest rung — no objectives, just the model.
A model dimension of 256: a 2048-unit recurrence in eight blocks, 16 classes per latent, and a first convolution of 16 channels. This is the rung the paper uses for both Control Suites, where it reports the same performance as the 200M model while being substantially faster.
Model Size
Parameters
pretrainedbool= FalseNo weights are published for this family; passing
True raises.**overridesobject= {}Optional
DreamerV3Config field overrides. action_dim
and action_space are set by the environment, not by the rung.Returns
DreamerV3ModelEncoder, RSSM, decoder, reward head, both critics and the actor.
Notes
Reference: Hafner, Pasukonis, Ba, and Norouzi, "Mastering Diverse Domains through World Models", Nature 640 (2025), 647-653 (arXiv:2301.04104), Table 3.
Examples
>>> from lucid.models import dreamer_v3_12m
>>> model = dreamer_v3_12m(action_dim=6).eval()
>>> model.config.deter_size, model.config.discrete, model.config.blocks
(2048, 16, 8)