hongshaorou/dreamer4
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

they decided on 0.3 for the behavioral prior loss weight

Browse Source
This commit is contained in:
lucidrains 2025-10-29 13:24:58 -07:00
parent 35b87c4fa1
commit 767789d0ca
2 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
dreamer4/dreamer4.py
View File

@ -1900,7 +1900,7 @@ class DynamicsWorldModel(Module):
gae_lambda = 0.95, gae_lambda = 0.95,
ppo_eps_clip = 0.2, ppo_eps_clip = 0.2,
pmpo_pos_to_neg_weight = 0.5, # pos and neg equal weight pmpo_pos_to_neg_weight = 0.5, # pos and neg equal weight
pmpo_kl_div_loss_weight = 1., pmpo_kl_div_loss_weight = .3,
value_clip = 0.4, value_clip = 0.4,
policy_entropy_weight = .01, policy_entropy_weight = .01,
gae_use_accelerated = False gae_use_accelerated = False

2
pyproject.toml
View File

@ -1,6 +1,6 @@
[project] [project]
name = "dreamer4" name = "dreamer4"
version = "0.0.99" version = "0.0.100"
description = "Dreamer 4" description = "Dreamer 4"
authors = [ authors = [
{ name = "Phil Wang", email = "lucidrains@gmail.com" } { name = "Phil Wang", email = "lucidrains@gmail.com" }