Implementation:Bentoml BentoML Models Export Import

Implementation Metadata
Implementation Name	Models Export Import
API	bentoml.models.export_model(), bentoml.models.import_model()
Source	src/bentoml/models.py:L62-207
Workflow	Model_Store_Management
Domain	ML_Serving, Model_Management
Implements	Principle:Bentoml_BentoML_Model_Export_Import
Last Updated	2026-02-13 15:00 GMT

Overview

The bentoml.models.export_model() and bentoml.models.import_model() functions provide portable serialization and deserialization of BentoML models. Export packages a model from the local store into an archive file (or folder), while import reconstructs a model from such an archive into the local store.

Import

import bentoml

Signatures

export_model()

def export_model(
    tag: Tag | str,
    path: str,
    output_format: str = None,
    *,
    protocol: str = None,
    user: str = None,
    passwd: str = None,
    params: dict = None,
    subpath: str = None,
) -> str

import_model()

def import_model(
    path: str,
    input_format: str = None,
    *,
    protocol: str = None,
    user: str = None,
    passwd: str = None,
    params: dict = None,
    subpath: str = None,
) -> Model

Parameters

export_model() Parameters

Parameter	Type	Default	Description
tag	str	required	The model tag to export from the local store.
path	str	required	Destination path. Can be a local filesystem path or a remote URI (e.g., s3://bucket/path).
output_format	str	None	Archive format. If None, inferred from the file extension. Supported: "bentomodel", "gz", "xz", "bz2", "tar", "zip", "folder".
protocol	str	None	Remote protocol override (e.g., "s3", "ftp"). Usually inferred from the URI scheme.
user	str	None	Username for remote protocol authentication.
passwd	str	None	Password for remote protocol authentication.
params	dict	None	Additional parameters for the fsspec filesystem backend.
subpath	str	None	Subpath within the remote filesystem.

import_model() Parameters

Parameter	Type	Default	Description
path	str	required	Source path to import from. Can be local or remote.
input_format	str	None	Archive format. If None, inferred from file extension.
protocol	str	None	Remote protocol override.
user	str	None	Username for authentication.
passwd	str	None	Password for authentication.
params	dict	None	Additional fsspec parameters.
subpath	str	None	Subpath within the remote filesystem.

Inputs and Outputs

export_model()

Inputs:

• Model tag (must exist in local store) + destination path

Outputs:

• str — the path where the model was exported

import_model()

Inputs:

• Source path (local file, remote URI, or folder)

Outputs:

• Model — the imported model, now available in the local store

Supported Formats

Format String	File Extension	Description
"bentomodel"	.bentomodel	Default BentoML archive format
"gz"	.gz	Gzip-compressed tar
"xz"	.xz	XZ-compressed tar
"bz2"	.bz2	Bzip2-compressed tar
"tar"	.tar	Uncompressed tar
"zip"	.zip	ZIP archive
"folder"	(directory)	Direct folder copy

Usage Examples

import bentoml

# Export to a local file (format inferred from extension)
path = bentoml.models.export_model(
    "text_classifier:latest",
    "/tmp/text_classifier.bentomodel",
)
print(f"Exported to: {path}")

# Export with explicit gzip compression
bentoml.models.export_model(
    "text_classifier:latest",
    "/tmp/text_classifier.tar.gz",
    output_format="gz",
)

# Export to S3
bentoml.models.export_model(
    "text_classifier:latest",
    "s3://my-bucket/models/text_classifier.bentomodel",
)

# Import from a local file
model = bentoml.models.import_model("/tmp/text_classifier.bentomodel")
print(f"Imported: {model.tag}")

# Import from S3
model = bentoml.models.import_model(
    "s3://my-bucket/models/text_classifier.bentomodel"
)

Behavior Details

• Format is auto-detected from the file extension when output_format / input_format is None.
• Remote protocol support is powered by fsspec, enabling seamless integration with S3, GCS, FTP, HDFS, and other backends.
• The exported archive is self-contained, including the model.yaml descriptor, all artifact files, and custom objects.
• On import, the model is added to the local store with its original tag. If a model with the same tag already exists, the import will fail unless the existing model is deleted first.

Source Reference

File: src/bentoml/models.py, lines 62-207.

Knowledge Sources

• BentoML
• BentoML Docs