Module checkpoint

Expand description

Checkpoint — save and load model parameters.

Structs§

TrainingCheckpoint: Complete training checkpoint: model weights + optimizer state + training metadata.

from_bytes: Deserialize named tensors from an in-memory byte slice.
load: Load parameters from a checkpoint file into an Executor.
load_tensors: Load named tensors from a file.
load_trainer: Load parameters from a checkpoint file into a Trainer.
load_training: Load a complete training checkpoint from a file.
read_checkpoint: Read named tensors from a reader in the Shrew checkpoint format.
read_training_checkpoint: Read a full training checkpoint from a reader.
save: Save all parameters from an Executor to a checkpoint file.
save_tensors: Save a list of named tensors to a file.
save_trainer: Save all parameters from a Trainer to a checkpoint file.
save_training: Save a complete training checkpoint to a file.
to_bytes: Serialize named tensors to an in-memory byte vector.
training_from_bytes: Deserialize a training checkpoint from an in-memory byte slice.
training_to_bytes: Serialize a training checkpoint to an in-memory byte vector.
write_checkpoint: Write a set of named tensors to a writer in the Shrew checkpoint format.
write_training_checkpoint: Write a full training checkpoint to a writer.