optscale-arcee 0.1.36

ML profiling tool for OptScale

Arcee

The OptScale ML profiling tool by Hystax

Arcee is a tool that hepls you to integrate ML tasks with OptScale. This tool can automatically collect executor metadata from cloud and process stats.

Installation

Arcee requires python 3.7+ to run.

pip install optscale-arcee

Usage

First of all you need to import and init arcee in your code:

import optscale_arcee as arcee

# init arcee using context manager syntax
with arcee.init('token', 'model_key'):
    # some code

To use custom endpoint and enable\disable ssl checks (supports using self-signed ssl certificates):

with arcee.init('token', 'model_key', endpoint_url='https://my.custom.endpoint:443/arcee/v2', ssl=False):
    # some code

Alternatively arcee can be initialized via function call. However manual finish is required:

arcee.init('token', 'model_key')
# some code
arcee.finish()

Or in error case:

arcee.init('token', 'model_key')
# some code
arcee.error()

To send stats:

arcee.send({"loss": 2.0012, "iter": 2, "epoch": 1})

(key should be string, value - int or float, multiple values can be sent)

To add tags to model run (key, value):

arcee.tag("project", "torchvision demo")

To add milestones:

arcee.milestone("Download test data")

To add stages:

arcee.stage("calculation")