dreamer_v3_200m(pretrained: bool = False, overrides: object = {})Construct the paper's default rung — no objectives, just the model.
A model dimension of 1024: an 8192-unit recurrence in eight blocks, 64 classes per latent, 64 channels at the first convolution. This is the size behind every headline result in the paper except the two Control Suites — Minecraft, DMLab, ProcGen, Atari, Atari100K and BSuite all run here, with identical hyperparameters.
Model Size
Parameters
pretrainedbool= FalseNo weights are published for this family; passing
True raises.**overridesobject= {}Optional
DreamerV3Config field overrides.Returns
DreamerV3ModelThe trunk at 200M parameters.
Notes
Reference: Hafner, Pasukonis, Ba, and Norouzi, "Mastering Diverse Domains through World Models", Nature 640 (2025), 647-653 (arXiv:2301.04104), Tables 2 and 3.
Examples
>>> from lucid.models import dreamer_v3_200m
>>> model = dreamer_v3_200m(action_dim=18, action_space="discrete").eval()
>>> model.config.deter_size, model.config.discrete
(8192, 64)