Fix: explicitly cast to bool tensor before masked scatter

modeling_deepseekocr2.py

@@ -497,7 +497,10 @@ class DeepseekOCR2Model(DeepseekV2Model):
                         device = torch.device("cuda")
                     else:
                         device = torch.device("cpu")
-                    inputs_embeds[idx].masked_scatter_(images_seq_mask[idx].unsqueeze(-1).to(device), images_in_this_batch)
+                        
+                    mask = images_seq_mask[idx].unsqueeze(-1).to(device=device, dtype=torch.bool)
+    
+                    inputs_embeds[idx].masked_scatter_(mask, images_in_this_batch)
 
                 idx += 1