Update modeling_minicpmv.py

modeling_minicpmv.py

@@ -308,11 +308,11 @@ class MiniCPMV(MiniCPMVPreTrainedModel):
             batch[place] = tensor
             attention_mask.append(to_add)
         
-        # attention_mask = torch.tensor(attention_mask)
+        attention_mask = torch.tensor(attention_mask)
         
         print(attention_mask.shape)
         print(attention_mask, "ATTENTION")
-        attention_mask = attention_mask.to(self.device)
+        # attention_mask = attention_mask.to(self.device)
         # padded_tensors = [torch.nn.functional.pad(tensor, (0, 0, 0, max_x - tensor.shape[1])) for tensor in batch]
         
         # Step 3: Stack the padded tensors into a single batch