offline did not use cosine learning rate decay

quasimetric-learning · Aug 11, 2023 · 4f11323 · 4f11323
1 parent f3cce80
commit 4f11323
Showing 1 changed file with 0 additions and 2 deletions.
diff --git a/offline/run_maze2d.sh b/offline/run_maze2d.sh
@@ -15,13 +15,11 @@ args=(
     agent.quasimetric_critic.losses.latent_dynamics.weight=1
     # critic
     agent.quasimetric_critic.losses.critic_optim.lr=5e-4
-    agent.quasimetric_critic.losses.critic_optim.cosine_lr_decay_final_mul=0
     agent.quasimetric_critic.losses.global_push.softplus_beta=0.01
     agent.quasimetric_critic.losses.global_push.softplus_offset=500
     # actor
     agent.actor.model.arch="[1024,1024,1024,1024]"
     agent.actor.losses.actor_optim.lr=3e-5
-    agent.actor.losses.actor_optim.cosine_lr_decay_final_mul=0
     agent.actor.losses.min_dist.adaptive_entropy_regularizer=False
     agent.actor.losses.min_dist.add_goal_as_future_state=False
     agent.actor.losses.behavior_cloning.weight=0.05