Struct CheckpointManager 

pub struct CheckpointManager { /* private fields */ }

Gradient checkpointing manager

Implementations

impl CheckpointManager

pub fn new(config: CheckpointConfig) -> Result<Self, OptimizerError>

Create a new checkpoint manager

pub fn should_checkpoint(&self) -> bool

Check if a checkpoint should be saved at the current step

pub fn save_checkpoint( &mut self, optimizer_state: Option<&[u8]>, gradients: Option<&HashMap<String, Vec<f32>>>, model_parameters: Option<&HashMap<String, Vec<f32>>>, loss: Option<f32>, learning_rate: f32, epoch: Option<usize>, ) -> Result<PathBuf, OptimizerError>

Save a checkpoint

pub fn load_checkpoint( &self, checkpoint_path: &Path, ) -> Result<Checkpoint, OptimizerError>

Load a checkpoint

pub fn latest_checkpoint(&self) -> Option<&PathBuf>

Get the latest checkpoint path

pub fn list_checkpoints(&self) -> Vec<&CheckpointMetadata>

Get all checkpoint metadata

pub fn step(&mut self)

Increment current step

pub fn current_step(&self) -> usize

Get current step

pub fn set_step(&mut self, step: usize)

Set current step

pub fn resume_from_latest( &mut self, ) -> Result<Option<Checkpoint>, OptimizerError>

Resume from the latest checkpoint

pub fn cleanup(&mut self) -> Result<(), OptimizerError>

Clean up checkpoints manually

pub fn statistics(&self) -> CheckpointStatistics

Get checkpoint statistics