Update modeling_quiet.py

modeling_quiet.py

@@ -1454,10 +1454,11 @@ class QuietForCausalLM(QuietPreTrainedModel, GenerationMixin):
 		inputs_embeds: Optional[torch.FloatTensor] = None,
 		labels: Optional[torch.LongTensor] = None,
 		use_cache: Optional[bool] = None,
-		# output_router_logits: Optional[bool] = None,
 		output_attentions: Optional[bool] = None,
 		output_hidden_states: Optional[bool] = None,
 		return_dict: Optional[bool] = None,
+		max_new_tokens: Optional[int] = None,
+		temperature: Optional[float] = None,
 		temperature_last: Optional[float] = None,
 		dynamic_temperature: Optional[float] = None,
 		dynatemp_low: Optional[float] = None,
@@ -1483,6 +1484,13 @@ class QuietForCausalLM(QuietPreTrainedModel, GenerationMixin):
 		do_sample: Optional[bool] = None,
 		encoder_repetition_penalty: Optional[float] = None,
 		no_repeat_ngram_size: Optional[int] = None,
+		sampler_priority: Optional[List[str]] = None,
+		negative_prompt_ids: Optional[List[int]] = None,
+		prompt_lookup_num_tokens: Optional[int] = None,
+		epsilon_cutoff: Optional[float] = None,
+		eta_cutoff: Optional[float] = None,
+		suppress_tokens: Optional[List[int]] = None,
+		synced_gpus: Optional[bool] = None,
 	) -> Union[Tuple, CausalLMOutputWithPast]:
 		r"""
 		Args: