QuickStart

In this Notebook we run Archai’s Quickstart example on Azure Machine Learning.

Prerequisites

• Python 3.7 or later

• An Azure subscription

• An Azure Resource Group

• An Azure Machine Learning Workspace

This notebook also assumes you have a python environment setup using pip install -e .[aml] in your Archai repository root

from pathlib import Path

from IPython.display import display, Image
from IPython.core.display import HTML

from azure.ai.ml import Output, command

import archai.common.azureml_helper as aml_helper
import archai.common.notebook_helper as nb_helper

Get a handle to the workspace

We load the workspace from a workspace configuration file.

ml_client = aml_helper.get_aml_client_from_file("../.azureml/config.json")
print(f'Using workspace: {ml_client.workspace_name} in resource group: {ml_client.resource_group_name}')

Create a compute cluster

We provision a Linux compute cluster for this Notebook. See the full list on VM sizes and prices.

cpu_compute_name = "nas-cpu-cluster-D14-v2"
compute_cluster = aml_helper.create_compute_cluster(ml_client, cpu_compute_name)

You already have a cluster named nas-cpu-cluster-D14-v2, we'll reuse it as is.

Create an environment based on a YAML file

Azure Machine Learning maintains a set of CPU and GPU Ubuntu Linux-based base images with common system dependencies. For the set of base images and their corresponding Dockerfiles, see the AzureML Containers repo.

archai_job_env = aml_helper.create_environment_from_file(ml_client,
                                                         image="mcr.microsoft.com/azureml/openmpi3.1.2-ubuntu18.04:latest",
                                                         conda_file="conda.yaml",
                                                         version="0.0.1")

Environment with name aml-archai is registered to workspace, the environment version is 0.0.1

Create job

job = command(experiment_name="archai_quickstart",
              display_name="Archai's QuickStart",
              compute=cpu_compute_name,
              environment=f"{archai_job_env.name}:{archai_job_env.version}",
              code="main.py",
              outputs=dict(
                  output_path=Output(type="uri_folder", mode="rw_mount")
              ),
              command="python main.py --output_dir ${{outputs.output_path}}"
              )

Run job

quickstart_job = ml_client.create_or_update(job)

Uploading main.py (< 1 MB): 100%|##########| 1.74k/1.74k [00:00<00:00, 6.91kB/s]

Open the job overview on Azure ML Studio in your web browser (this works when you are running this notebook in VS code).

import webbrowser
webbrowser.open(quickstart_job.services["Studio"].endpoint)

job_name = quickstart_job.name
print(f'Started job: {job_name}')

Started job: busy_shampoo_cqjgwy28gc

Download job’s output

output_name = "output_path"
download_path = "output"

aml_helper.download_job_output(ml_client, job_name=quickstart_job.name, output_name=output_name, download_path=download_path)

downloaded_folder = Path(download_path) / "named-outputs" / output_name

Show Pareto Frontiers

param_vs_latency_img = Image(filename=downloaded_folder / "pareto_non_embedding_params_vs_onnx_latency.png")
display(param_vs_latency_img)

param_vs_memory_img = Image(filename=downloaded_folder / "pareto_non_embedding_params_vs_onnx_memory.png")
display(param_vs_memory_img)

latency_vs_memory_img = Image(filename=downloaded_folder / "pareto_onnx_latency_vs_onnx_memory.png")
display(latency_vs_memory_img)

Show search state of the last iteration

df = nb_helper.get_search_csv(downloaded_folder)
df = df[['archid', 'non_embedding_params', 'onnx_latency', 'onnx_memory', 'is_pareto']]
df[(df['onnx_latency'] < 0.1) & (df['is_pareto'] == True)]

	archid	non_embedding_params	onnx_latency	onnx_memory	is_pareto
1	gpt2_2ce523ffef6587e0a9790c173a89d6fd25f1b9b6	7769472.0	0.068997	54.867023	True
5	gpt2_e24cd64a53d6a9be4a5cf2116daee38e0763947c	9791616.0	0.087752	57.236445	True
7	gpt2_ad522de7b54d7ad73231ccab52e7f8b17c444dd8	3154688.0	0.027303	45.336156	True
8	gpt2_3b76e4046ead432cab16d17d0a3d500efc0249c5	1963840.0	0.025981	13.966668	True
9	gpt2_56b3029ce65d7c75585e97c67a3162b6b34fd453	9462656.0	0.077058	61.318761	True
12	gpt2_af14be0b4e8f8a29e744cd81c503419200e05c3a	2564480.0	0.024791	43.084689	True
13	gpt2_08a67612ee38cccd05c47a45e137a8e82921c3c0	5356160.0	0.055600	34.976172	True
15	gpt2_1638d1cbba003004298c9f1fec1885e1c3e724ac	1841408.0	0.015650	29.566623	True
17	gpt2_a9736f6b29e56c1026b452ede20adc7b9f488ada	10346304.0	0.077892	72.751252	True
19	gpt2_35dc3c85cfeb893d23399a0d0c087ae53f718323	1782528.0	0.018778	15.950404	True
23	gpt2_49d99f06ad70afece175610b82e4b9f69a367d07	1185408.0	0.015044	27.064179	True
29	gpt2_6dcd8caf6477c3554cefa5a9629f2778e31582e8	6409472.0	0.054197	52.364760	True
31	gpt2_adcecc4a73970b5ceaf61c5a20c6e2fc914bebbf	1683328.0	0.018393	12.884486	True
33	gpt2_ea6fcf50e01df27f5065286f64d3d4fd086ad1f6	6908352.0	0.062050	40.857306	True
34	gpt2_3df377a7c2f287c0396798d69587be4ce80eb065	9476096.0	0.082323	58.701959	True
35	gpt2_5825b8567da483715267609a45897741217c7358	9194496.0	0.069051	73.742070	True
38	gpt2_1b55c333a06a31b8c3f5dee0be477bff60961b27	7109376.0	0.062521	44.316129	True
42	gpt2_6f1494e13898235ea0abfee60f651e972ad7c15f	8138432.0	0.071064	56.273517	True
44	gpt2_05bb7d11f19a7864283f4a4c48d205a71bbc0cca	5295232.0	0.048272	61.571509	True
45	gpt2_88d96623f6fba1aef736f1eebfde662ebd2dc91c	10352192.0	0.086744	64.712472	True
52	gpt2_cc745fafec97887cfbd9583cd89c276805121d14	8575680.0	0.069086	65.999776	True
53	gpt2_54a4da9b9011aa8b7cabd16c81d7ec60e1fa4bec	6503936.0	0.052899	55.414674	True
62	gpt2_1c4ef81c171f066ea6e51256051fb83af5ce9ee3	12931200.0	0.098794	90.671401	True