topic Re: How can I view the storage space taken by a registered model using MLFlow? in Machine Learning

How can I view the storage space taken by a registered model using MLFlow?

fa — Thu, 08 Dec 2022 16:34:09 GMT

The information viewed about the registered models on the Models tab is very minimal. Just showing the tags we pass in and version information. How can I get more details about the model such as the size on disk?

Re: How can I view the storage space taken by a registered model using MLFlow?

Wunlung — Thu, 08 Dec 2022 23:41:45 GMT

The registered model actually links with a source run in the mlflow experiment and get those information from that experiment.

A. Using the UI.

Select the model under Registered Models
Select the version which you want to get more detail
Click the link Near the Source Run and redirect you to the experiment page
You will get all the artifacts including model object(size, path, ...), input_example.json, yaml, requirement.txt, ...

B. Use the MLFlowClient

from pprint import pprint
 
client = MlflowClient()
for rm in client.search_registered_models():
    pprint(dict(rm), indent=4)

You will get the source path and can list those objects.

Re: How can I view the storage space taken by a registered model using MLFlow?

Octavian1 — Tue, 13 Feb 2024 09:58:41 GMT

Hi,

I have used the MLFlow client, but I am not sure where to find the size of the model image.
The response to

client.search_registered_models()

I am getting is the following:

<RegisteredModel: aliases={}, creation_timestamp=17061..., description='', last_updated_timestamp=17061..., latest_versions=None, name='<catalog>.<db>.<model>', tags={}>
{ 'aliases': {},
'creation_timestamp': 17078...,
'description': '',
'last_updated_timestamp': 17078...,
'latest_versions': None,
'name': '<catalog>.<db>.<model>',
'tags': {}}

If going to the catalog UI view, where the model and its versions details are given, there is no such information.