Update.

diff --git a/tasks.py b/tasks.py
index cdf8f9e..b967465 100755 (executable)
--- a/tasks.py
+++ b/tasks.py
@@ -276,6 +276,9 @@ class World(Task):
 
         average_logits = sum_logits / quizzes.numel()
 
+        # It's a bit brutal to do it twice, we should probably have a
+        # moving average and apply it right away
+
         if desired_average_logits is not None:
             temperature = average_logits / desired_average_logits
             masked_inplace_autoregression(