Implementation:Pytorch Serve ModelArchiverConfig

Overview

Pytorch_Serve_ModelArchiverConfig is the configuration dataclass that defines all parameters required for creating a Model Archive (MAR) file. It centralizes the archiver's configuration into a single, typed structure using Python's @dataclass decorator.

Source

Property	Value
File	`model-archiver/model_archiver/model_archiver_config.py`
Lines	28
Language	Python
Key Class	`ModelArchiverConfig` (lines 9–28)

Class Definition

The ModelArchiverConfig class is decorated with @dataclass and contains all fields necessary to drive the model archiving process.

Fields

Field	Type	Default	Description
`model_name`	`str`	required	Name of the model to archive
`handler`	`str`	required	Handler file or built-in handler name
`version`	`str`	required	Model version string
`serialized_file`	`Optional[str]`	`None`	Path to the serialized model weights file (e.g., `.pt`, `.pth`)
`model_file`	`Optional[str]`	`None`	Path to the model architecture definition file
`extra_files`	`Optional[str]`	`None`	Comma-separated list of additional files to include in the archive
`runtime`	`str`	`"PYTHON"`	Runtime environment for the model
`export_path`	`str`	current working directory	Directory where the archive will be written
`archive_format`	`Literal["default", "tgz", "no-archive"]`	N/A	Archive output format
`force`	`bool`	N/A	Whether to overwrite an existing archive
`requirements_file`	`Optional[str]`	`None`	Path to a pip requirements file to bundle
`config_file`	`Optional[str]`	`None`	Path to a model-specific config file

Class Method: from_args

@classmethod
def from_args(cls, args: Namespace) -> "ModelArchiverConfig":
    ...

The from_args class method constructs a ModelArchiverConfig instance from an argparse.Namespace object. This bridges the CLI argument parsing layer with the typed configuration layer, allowing the archiver to be invoked both programmatically and from the command line.

Usage Pattern

from model_archiver.model_archiver_config import ModelArchiverConfig

# From parsed CLI arguments
config = ModelArchiverConfig.from_args(parsed_args)

# Or directly via dataclass construction
config = ModelArchiverConfig(
    model_name="resnet-18",
    handler="image_classifier",
    version="1.0",
    serialized_file="resnet18.pt",
    archive_format="default",
    force=True,
)

Relationship to Principles

This implementation directly supports the Pytorch_Serve_Model_Archiving principle. The ModelArchiverConfig dataclass encapsulates every parameter the archiver needs, ensuring that the packaging of models into MAR files is driven by a well-defined, validated configuration object.

Metadata

Pytorch_Serve Pytorch_Serve_Model_Archiving 2026-02-13 18:52 GMT

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment