Fixing the multi gpu inference bug

modeling_internlm_xcomposer2.py

@@ -185,7 +185,7 @@ class InternLMXComposer2ForCausalLM(InternLM2PreTrainedModel):
                 wrap_im_mask.append(torch.zeros(part_embeds.shape[:2]))
                 temp_len += part_embeds.shape[1]
             if idx < image_nums:
-                wrap_embeds.append(image[idx].unsqueeze(0))
+                wrap_embeds.append(image[idx].unsqueeze(0).to(self.device))
                 wrap_im_mask.append(torch.ones(1, image[idx].shape[0]))
                 temp_len += im_len