wenmeng zhou

we will refine this in next version

welcome to create a pull request to fix this for ARM

```shell 2023-03-19T06:59:14.5461538Z 2023-03-19T06:59:14.5461868Z Case pipelines.test_image_open_vocabulary_detection.ImageOpenVocabularyDetectionTest.test_run_with_model_from_modelhub run result: Error, msg: 2023-03-19T06:59:14.5462323Z Traceback (most recent call last): 2023-03-19T06:59:14.5462849Z File "/opt/conda/lib/python3.7/site-packages/tensorflow_core/python/framework/importer.py", line 501, in _import_graph_def_internal 2023-03-19T06:59:14.5463400Z graph._c_graph, serialized, options) # pylint: disable=protected-access 2023-03-19T06:59:14.5465117Z...

thanks， can you open a pull request to fix this issue?

the related commit has been reverted: https://github.com/modelscope/modelscope/commit/77ff1d94d646cee88457419512ee55a3152b3db4 https://github.com/modelscope/modelscope/commit/f2f48dd30565436db25bbeb05e07930a6038469c

pip install control_ldm -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html

参考这里： https://github.com/open-mmlab/mmcv#installation

control_type='scribble' 在构造函数中传递(构建pipeline时)，而不是__call_函数中传递

1.3.0 版本中 tensor 并行训练过程中每次保存 ckpt 都会保存对应训练线程数的 pth 文件，文件名中带有 rank_0x 信息，将这些 pth 文件重命名为 mp_rank_0x_model_states.pt 放入 /model 路径下即可使用

> 1.3版本测试了，还是报上面的错误 It has been fixed in version 1.4