finetune

modeling_chatglm.py

@@ -884,6 +884,9 @@ class ChatGLMModel(ChatGLMPreTrainedModel):
 
         batch_size, seq_length = input_ids.shape
 
+        if inputs_embeds is None:
+            inputs_embeds = self.embedding(input_ids)
+
         if self.pre_seq_len is not None:
             if past_key_values is None:
                 past_key_values = self.get_prompt(batch_size=batch_size, device=input_ids.device,
@@ -912,9 +915,8 @@ class ChatGLMModel(ChatGLMPreTrainedModel):
 
                     attention_mask = torch.stack(new_attention_mask, dim=0)
                     input_ids = torch.stack(new_input_ids, dim=0)
+                    inputs_embeds = self.embedding(input_ids)
 
-                if inputs_embeds is None:
-                        inputs_embeds = self.embedding(input_ids)
                 full_attention_mask = self.get_masks(inputs_embeds, past_key_values, padding_mask=attention_mask)
 
         # Rotary positional embeddings