Fix SFT missing arg

Signed-off-by: jiemingz <jiemingz@nvidia.com>

nemo/collections/nlp/models/language_modeling/megatron_gpt_sft_model.py

@@ -320,7 +320,7 @@ def _determine_log_key(self, data_config, dataloader_idx, metric_name, mode):
  else:
  return base_key + f"dataloader{dataloader_idx}"
 
- def fwd_bwd_step(self, dataloader_iter, batch_idx, forward_only):
+ def fwd_bwd_step(self, dataloader_iter, batch_idx, forward_only, first_val_step=None):
  batch = next(dataloader_iter)
 
  log_token_counts = self.cfg.get('log_token_counts', False)
@@ -360,6 +360,7 @@ def fwd_bwd_step(self, dataloader_iter, batch_idx, forward_only):
  forward_only=forward_only,
  seq_length=seq_length,
  micro_batch_size=get_micro_batch_size(),
+ first_val_step=first_val_step,
  )
 
  # only the last stages of the pipeline return losses