mapping_networks.checkpoint.CheckpointSchema¶

class mapping_networks.checkpoint.CheckpointSchema(schema_version: int, package_version: str, latent_state: dict[str, ~typing.Any], mapper_buffers: dict[str, ~typing.Any], parameter_spec_names: list[tuple[str, list[int]]], config: dict[str, ~typing.Any] | None = None, trainer_state: dict[str, ~typing.Any] | None = None, metadata: dict[str, ~typing.Any] = <factory>)¶

All fields written to / read from a checkpoint file.

schema_version¶

Integer incremented on breaking format changes.

Type:: int

package_version¶

mapping_networks.__version__ at save time.

Type:: str

latent_state¶

Ordered state_dict of trainable latent tensors only.

Type:: dict[str, Any]

mapper_buffers¶

Buffer state_dict required for deterministic reconstruction (projection matrices, running statistics …).

Type:: dict[str, Any]

parameter_spec_names¶

Ordered list of (name, shape) tuples from the target ParameterSpec. Validated on load.

Type:: list[tuple[str, list[int]]]

config¶

Optional serialised MappingConfig dict (may be None).

Type:: dict[str, Any] | None

trainer_state¶

Optional dict with optimizer/scheduler/epoch entries.

Type:: dict[str, Any] | None

metadata¶

User-supplied annotations, e.g. dataset or description.

Type:: dict[str, Any]

__init__(schema_version: int, package_version: str, latent_state: dict[str, ~typing.Any], mapper_buffers: dict[str, ~typing.Any], parameter_spec_names: list[tuple[str, list[int]]], config: dict[str, ~typing.Any] | None = None, trainer_state: dict[str, ~typing.Any] | None = None, metadata: dict[str, ~typing.Any] = <factory>) → None¶

Methods

__init__(schema_version, package_version, ...)

Attributes

`config`
`trainer_state`
`schema_version`
`package_version`
`latent_state`
`mapper_buffers`
`parameter_spec_names`
`metadata`