Add ignore_index in DPOTrainer's nn.CrossEntropyLoss (#1987)

Co-authored-by: Kashif Rasul <kashif.rasul@gmail.com>

trl/trainer/dpo_trainer.py

@@ -1401,7 +1401,7 @@ def cross_entropy_loss(logits, labels):
                 logits = logits[..., :-1, :].contiguous()
                 labels = labels[..., 1:].contiguous()
             # Flatten the tokens
-            loss_fct = nn.CrossEntropyLoss()
+            loss_fct = nn.CrossEntropyLoss(ignore_index=self.label_pad_token_id)
             logits = logits.view(-1, logits.shape[-1])
             labels = labels.view(-1)
             # Enable model parallelism