MLEM now offers deployment to Kubernetes and Sagemaker with a single command.
MLEM objects (models, data, etc.) can live in different locations such as Git repositories, cloud object storage, local directories, etc.
In this page we will work with model objects in Git repos, but the same operations apply to any object type and location.
Although you can store MLEM objects in any location such as a Git repo, Cloud storage, or external drives, creating a MLEM project lets you organize MLEM objects consistently.
To create a MLEM project in a remote location, you can provide its URL or path
to mlem init.
You can load MLEM objects from remote locations inside Python code with
mlem.api.load() by using an object name and its URL.
from mlem.api import load
model = load(
"models/rf",
project="https://github.com/iterative/example-mlem-get-started",
rev="main"
)This fetches the rf model from branch main of the
example-mlem-get-started repo and loads it to memory.
You can download MLEM objects to the local environment in with mlem clone
(CLI).
$ mlem clone models/rf \
--project https://github.com/iterative/example-mlem-get-started \
ml_model
⏳️ Loading meta from https://github.com/iterative/example-mlem-get-started/tree/main/models/rf.mlem
🐏 Cloning https://github.com/iterative/example-mlem-get-started/tree/main/models/rf.mlem
💾 Saving model to ml_model.mlemThis copies the rf model from branch main of the
example-mlem-get-started repo to the current directory and renames it to
ml_model.
It's also possible to (down)load loose MLEM objects stored in any cloud
platform supported by fsspec, e.g. Amazon S3. To do so, provide the file
system protocol & path as target/URL, e.g. s3://<bucket>/
Loose objects are typically stored this way because they do not require versioning.
$ mlem clone rf s3://example-mlem-get-started/rf
⏳️ Loading meta from rf.mlem
🐏 Cloning rf.mlem
💾 Saving model to s3://example-mlem-get-started/rf.mlemThe rf model from S3 bucket example-mlem-get-started can also be
loaded via API or used in the CLI as though it existed
locally:
from mlem.api import load
model = load("rf", project="s3://example-mlem-get-started")$ mlem apply rf \
--project s3://example-mlem-get-started \
test_x.csv --json
[1, 0, 2, 1, 1, 0, 1, 2, 1, 1, 2, 0, 0, 0, 0, 1, 2, 1, 1, 2, 0, 2, 0, 2, 2, 2, 2, 2, 0, 0, 0, 0, 1, 0, 0, 2, 1, 0]