dreamer_v2_dmc(pretrained: bool = False, overrides: object = {})Construct DreamerV2 as the paper configures it for the Control Suite.
A 200-unit deterministic state, the full divergence floored at one free nat, a continuous policy, and no discount head — these episodes end on a clock rather than on a failure, and there is nothing for one to predict.
Model Size
Parameters
pretrainedbool= FalseNo weights are published for this family; passing
True raises.**overridesobject= {}Optional
DreamerV2Config field overrides. action_dim
is set by the task.Returns
DreamerV2ModelThe trunk, configured for the Control Suite.
Notes
Reference: Hafner, Lillicrap, Norouzi, and Ba, "Mastering Atari with Discrete World Models", ICLR, 2021 (arXiv:2010.02193).
The released implementation splits this into dmc_vision and
dmc_proprio, which differ only in whether the encoder reads pixels
or joint angles. This family reads pixels, so they are one
configuration here.
Examples
>>> from lucid.models import dreamer_v2_dmc
>>> model = dreamer_v2_dmc(action_dim=6).eval()
>>> model.config.deter_size, model.config.pcont
(200, False)