Get desktop application:
View/edit binary Protocol Buffers messages
Checkpoint a collection of files saved by a task.
Used in:
, , , , , , ,ID of the task which generated this checkpoint.
ID of the allocation which generated this checkpoint.
UUID of the checkpoint.
Timestamp when the checkpoint was reported.
Dictionary of file paths to file sizes in bytes of all files in the checkpoint.
User defined metadata associated with the checkpoint.
The state of the underlying checkpoint.
Training-related data for this checkpoint.
Optional ID that describes where this checkpoint is stored. It will be null on every checkpoint pre 0.27.1. It can also be null when a user does not specify the storageID calling the report API themselves or when users don't provide a storage config to core_context.
CheckpointTrainingMetadata is specifically metadata about training.
Used in:
The ID of the trial that created this checkpoint.
The ID of the experiment that created this checkpoint.
The configuration of the experiment that created this checkpoint.
Hyperparameter values for the trial that created this checkpoint.
Training metrics reported at the same steps_completed as the checkpoint.
Validation metrics reported at the same steps_completed as the checkpoint.
Searcher metric (as specified by the expconf) at the same steps_completed of the checkpoint.
Request to change checkpoint database information.
Used in:
The uuid of the checkpoint.
Dictionary of file paths to file sizes in bytes of all files in the checkpoint. This won't update actual checkpoint files. If len(resources) == 0 => the checkpoint is considered deleted Otherwise if resources are updated the checkpoint is considered partially deleted.
Gets around not being able to do "Optional map<string, int64>". Not ideal but this API is marked internal for now.
Used in:
Resources.
Sorts options for checkpoints by the given field.
Used in:
,Returns checkpoints in an unsorted list.
Returns checkpoints sorted by UUID.
Returns checkpoints sorted by trial id.
Returns checkpoints sorted by batch number.
Returns checkpoints sorted by end time.
Returns checkpoints sorted by state.
Returns checkpoints sorted by the experiment's `searcher.metric` configuration setting.
The current state of the checkpoint.
Used in:
, , ,The state of the checkpoint is unknown.
The checkpoint is in an active state.
The checkpoint is persisted to checkpoint storage.
The checkpoint errored.
The checkpoint has been deleted.
The checkpoint has been partially deleted.