AIExplainer
AI Hardware Intermediate 2 min read

What is checkpoint?

A snapshot of a model's state during training, allowing it to resume from that point if needed

A checkpoint is a way to save the progress of a machine learning model at a particular point in time, so it can be restarted from that point if something goes wrong or if the training process is interrupted

Think of a checkpoint like a save point in a video game, where you can save your progress and come back to it later if you need to

For example, if a researcher is training a large language model and the training process is interrupted due to a power outage, they can use a checkpoint to resume training from where they left off

Checkpoints are used to prevent losing progress in case of a failure, and to allow for the resumption of training from a previous point, which can save time and computational resources

Some people think that checkpoints are only used for backup purposes, but they can also be used to evaluate the performance of a model at different points during training

The concept of checkpoints has been around since the early days of machine learning, but it has become more important with the increasing complexity and size of modern models

save point snapshot model state

Three products for different needs — explore what’s relevant to you.