How Do I Use the Checkpoints of Built-in Algorithms to Train Models?
If you use the built-in algorithms of ModelArts to train models, you can use the checkpoints of the algorithms for training.
During model training, the TensorFlow-based built-in AI algorithm periodically saves the checkpoint file to the OBS path specified for Model Output. See Figure 1.
Find the output files in the corresponding path. The checkpoint file, .data, .index, and .meta files form a set of checkpoint. See Figure 2.
When the same algorithm is used for training again, the checkpoint of the existing training job can be copied to the model output path of the new training job. In this way, the new training job can be trained based on the existing checkpoint.


Did this article solve your problem?
Thank you for your score!Your feedback would help us improve the website.