Fix dialogue tutorial bug (NVIDIA#5297) (NVIDIA#5303)

* set add_pooling_layer=False for huggingface bert model

* remove add_pooling_layer=False and set find_unused_parameters=True

* set num_prompt_tokens to 0 for huggingface

Co-authored-by: Zhilin Wang <wangzhilin12061996@hotmail.com>
Co-authored-by: Eric Harper <complex451@gmail.com>

nemo/collections/nlp/models/dialogue/dialogue_gpt_classification_model.py

@@ -539,6 +539,7 @@ def generate_candidates(self, labels, template_length, input_ids, attn_masks):
                 for i in generated_tokens
             ]
             generated_tokens = torch.cat(generated_tokens, axis=0)
+            num_prompt_tokens = 0
 
         elif self.cfg.library == "megatron":