修复qwen2-vl-7b图像处理部分张量创建在cpu上导致报错问题

vllm/model_executor/models/qwen2_vl.py

@@ -746,7 +746,10 @@ class Qwen2VisionTransformer(nn.Module):
         rotary_pos_emb = self.rot_pos_emb(grid_thw)
 
         # compute cu_seqlens
-        grid_thw_ = torch.tensor(grid_thw)
+        grid_thw_ = torch.tensor(grid_thw,
+                                device=self.device,
+                                dtype=torch.int32
+                                )
         cu_seqlens = torch.repeat_interleave(grid_thw_[:, 1] * grid_thw_[:, 2],
                                              grid_thw_[:, 0]).cumsum(
                                                  dim=0, dtype=torch.int32)