Update modeling_quiet.py

modeling_quiet.py

@@ -1103,9 +1103,13 @@ class QuietForCausalLM(QuietPreTrainedModel):
         thought_ids = torch.zeros((batch_size, self.config.max_thoughts, max_length), dtype=torch.long, device=hidden_states.device)
         thought_embeddings = []
 
+        # Create an instance of QuietForCausalLM using the current model's configuration
+        causal_lm_model = QuietForCausalLM(self.config)
+        causal_lm_model.eval()  # Set the model to evaluation mode
+
         for i in range(self.config.max_thoughts):
             thought_input_ids = torch.zeros((batch_size, 1), dtype=torch.long, device=hidden_states.device)
-            thought_outputs = self.model.generate(
+            thought_outputs = causal_lm_model.generate(
                 input_ids=thought_input_ids,
                 max_length=max_length,
                 do_sample=True,
@@ -1115,12 +1119,13 @@ class QuietForCausalLM(QuietPreTrainedModel):
                 eos_token_id=self.config.eos_token_id,
             )
             thought_ids[:, i, :] = thought_outputs
-            thought_embeddings.append(self.model.get_input_embeddings()(thought_outputs))
+            thought_embeddings.append(self.get_input_embeddings()(thought_outputs))
 
         thought_embeddings = torch.stack(thought_embeddings, dim=1)
         return thought_ids, thought_embeddings
 
 
+
     def calculate_policy_loss(self, thoughts, rewards):
         thought_log_probs = []
         for thought in thoughts: