oops

Author: lucidrains

improving_transformers_world_model/agent.py

@@ -503,7 +503,7 @@ def critic_loss(
 
         # value clipping
 
-        clipped_value = old_values + (critic_value - old_values).clamp(1. - eps, 1. + eps)
+        clipped_value = old_values + (critic_value - old_values).clamp(-eps, eps)
 
         clipped_loss = loss_fn(clipped_value, returns, reduction = 'none')
         loss = loss_fn(critic_logits, returns, reduction = 'none')

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "improving-transformers-world-model"
-version = "0.0.55"
+version = "0.0.56"
 description = "Improving Transformers World Model for RL"
 authors = [
     { name = "Phil Wang", email = "lucidrains@gmail.com" }