diff --git a/nemo_aligner/models/nlp/gpt/megatron_gpt_dpo_model.py b/nemo_aligner/models/nlp/gpt/megatron_gpt_dpo_model.py
index 299959214..952b4e897 100644
--- a/nemo_aligner/models/nlp/gpt/megatron_gpt_dpo_model.py
+++ b/nemo_aligner/models/nlp/gpt/megatron_gpt_dpo_model.py
@@ -145,16 +145,6 @@ def fwd_output_and_loss_func(dataloader_iter, model, checkpoint_activations_all_
             # position_ids = batch["position_ids"][0:1]
             attention_mask = batch["attention_mask"][0:1]
 
-            pad_len = 0
-            world_size = torch.distributed.get_world_size()
-            if tokens.shape[1] % world_size != 0:
-                pad_len = world_size - tokens.shape[1] % world_size
-                with torch.no_grad():
-                    import torch.nn.functional as F
-
-                    tokens = F.pad(tokens, [0, pad_len, 0, 0], value=0)
-                    labels = F.pad(labels, [0, pad_len, 0, 0], value=0)
-
             # Model forward pass
             forward_args = {
                 "input_ids": tokens,