diff --git a/src/liger_kernel/chunked_loss/fused_linear_preference.py b/src/liger_kernel/chunked_loss/fused_linear_preference.py
index f59457c84..d970fcc5f 100644
--- a/src/liger_kernel/chunked_loss/fused_linear_preference.py
+++ b/src/liger_kernel/chunked_loss/fused_linear_preference.py
@@ -308,10 +308,6 @@ def chunk_forward(
                 reduction="sum",
                 ignore_index=ignore_index,
             )
-        else:
-            chosen_nll_loss = torch.zeros(
-                (), device=target_chunk.device, dtype=target_chunk.dtype
-            )
 
         loss_mask = target_chunk != ignore_index
         label_chunk = torch.where(loss_mask, target_chunk, 0)