Update modeling_quiet.py

modeling_quiet.py

@@ -2117,15 +2117,6 @@ class QuietForCausalLM(QuietPreTrainedModel, GenerationMixin):
 		del end_embedding
 		torch.cuda.empty_cache()
 
-		if streamer is not None:
-			streamer.put(
-				input_ids=input_ids[:, -1:], 
-				past_key_values=outputs.past_key_values,
-				hidden_states=outputs.hidden_states,
-				attentions=outputs.attentions,
-			)
-		
-
 		return CausalLMOutputWithPast(
 			loss=loss if loss is not None else None,
 			logits=(rm_logits if self.n_ahead > 1 else logits) if not self.output_logits_at_the_end else logits,