model checkpoint

A snapshot of model state saved periodically during training, containing key information such as model weights and optimizer state, used for failure recovery and subsequent fine-tuning.