lucidrains e2d86a4543 add a complete action embedder that can accept any number of discrete actions with variable bins as well as any number of continuous actions, pooled and added to the agent token as described in the paper (seems like they fixed that horrendous hack in dreamer v3 with sticky action)

2025-10-09 07:53:42 -07:00

..

__init__.py

for the value head, we will go for symexp encoding as well (following the "stop regressing" paper from Farebrother et al), also use layernormed mlp given recent papers

2025-10-08 07:37:34 -07:00

dreamer4.py

add a complete action embedder that can accept any number of discrete actions with variable bins as well as any number of continuous actions, pooled and added to the agent token as described in the paper (seems like they fixed that horrendous hack in dreamer v3 with sticky action)

2025-10-09 07:53:42 -07:00

trainers.py

add the noising of the latent context during generation, technique i think was from EPFL, or perhaps some google group that built on top of EPFL work

2025-10-07 09:37:37 -07:00