fix loss

turbo-llm · Nov 22, 2024 · 941dbe9 · 941dbe9
1 parent a9d9e6f
commit 941dbe9
Showing 1 changed file with 0 additions and 1 deletion.
diff --git a/turbo_alignment/trainers/dpo.py b/turbo_alignment/trainers/dpo.py
@@ -452,7 +452,6 @@ def compute_loss(
         rejected_rewards = self.beta * (policy_rejected_logps - reference_rejected_logps).detach()
 
         loss = -F.logsigmoid(self.beta * (logits - penalty_term))
-        loss = policy_chosen_logps
 
         return loss, chosen_rewards, rejected_rewards