Loading models

mlstac.load is the main entry point. It is an alias for ModelLoader, so both of these are equivalent:

import mlstac

model = mlstac.load("./my-model")
model = mlstac.ModelLoader("./my-model")

Accepted sources

load figures out how to reach the model from the value you pass in.

You pass	Detected as
https://..., http://..., ftp://...	remote HTTP(S)/FTP
s3://bucket/key	Amazon S3 (needs boto3)
gs://bucket/key	Google Cloud Storage (needs google-cloud-storage)
an existing local path	local files
a list of .pt2 files	an ad-hoc ensemble

# Remote metadata
mlstac.load("https://example.com/model/mlm.json")

# A local directory containing an mlm.json
mlstac.load("./my-model")

# An ensemble from .pt2 files
ensemble = mlstac.load(["model_a.pt2", "model_b.pt2"])
ensemble.is_ensemble  # True

Note

A bare name like "resnet50" is not a valid source on its own. Pass a full URL or a local path. If you pass an unresolvable name you get a clear error explaining what happened.

Inspecting a model

model = mlstac.load("./my-model")

model.get_model_summary()   # dict with id and key metadata
model.print_schema()        # human-readable schema
model.is_ensemble           # True for multi-model bundles