zishuo/onepanel
1
0
Fork 0
mirror of https://github.com/onepanelio/onepanel.git synced 2025-09-27 01:56:03 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

update MaskRCNN template

Browse Source
This commit is contained in:
rushtehrani
2021-01-29 12:07:25 -08:00
parent f1bdbbad28
commit 35ea2164eb
1 changed files with 124 additions and 118 deletions
Download Patch File Download Diff File Expand all files Collapse all files

242
db/yaml/workflows/hyperparameter-tuning/20210118175809.yaml
View File

@@ -1,11 +1,24 @@
# source: https://github.com/onepanelio/templates/blob/master/workflows/nni-hyperparameter-tuning/mnist/
# Workflow Template example for hyperparameter tuning
# Documentation: https://docs.onepanel.ai/docs/reference/workflows/hyperparameter-tuning
#
# Only change the fields marked with [CHANGE]
entrypoint: main
arguments:
parameters:
# [CHANGE] Path to your training/model architecture code repository
# Change this value and revision value to your code repository and branch respectively
- name: source
value: https://github.com/onepanelio/templates
# [CHANGE] Revision is the branch or tag that you want to use
# You can change this to any tag or branch name in your repository
- name: revision
value: master
value: v0.18.0
# [CHANGE] Default configuration for the NNI tuner
# See https://docs.onepanel.ai/docs/reference/workflows/hyperparameter-tuning#understanding-the-configurations
- name: config
displayName: Configuration
required: true
@@ -29,6 +42,9 @@ arguments:
command: python main.py --output /mnt/output
codeDir: .
# gpuNum: 1 # uncomment and update to number of GPUs
# [CHANGE] Search space configuration
# Change according to your hyperparameters and ranges
- name: search-space
displayName: Search space configuration
required: true
@@ -42,6 +58,9 @@ arguments:
"learning_rate": { "_type": "choice", "_value": [0.0001, 0.001, 0.01, 0.1] },
"epochs": { "_type": "choice", "_value": [10] }
}
# Node pool dropdown (Node group in EKS)
# You can add more of these if you have additional tasks that can run on different node pools
- displayName: Node pool
hint: Name of node pool or group to run this workflow task
type: select.nodepool
@@ -49,6 +68,110 @@ arguments:
value: {{.DefaultNodePoolOption}}
required: true
templates:
- name: main
dag:
tasks:
- name: hyperparameter-tuning
template: hyperparameter-tuning
- name: metrics-writer
template: metrics-writer
dependencies: [hyperparameter-tuning]
arguments:
# Use sys-metrics artifact output from hyperparameter-tuning Task
# This writes the best metrics to the Workflow
artifacts:
- name: sys-metrics
from: "{{tasks.hyperparameter-tuning.outputs.artifacts.sys-metrics}}"
- name: hyperparameter-tuning
inputs:
artifacts:
- name: src
# Clone the above repository into '/mnt/data/src'
# See https://docs.onepanel.ai/docs/reference/workflows/artifacts#git for private repositories
git:
repo: '{{workflow.parameters.source}}'
revision: '{{workflow.parameters.revision}}'
path: /mnt/data/src
# [CHANGE] Path where config.yaml will be generated or already exists
# Update the path below so that config.yaml is written to the same directory as your main.py file
# Note that your source code is cloned to /mnt/data/src
- name: config
path: /mnt/data/src/workflows/hyperparameter-tuning/mnist/config.yaml
raw:
data: '{{workflow.parameters.config}}'
# [CHANGE] Path where search_space.json will be generated or already exists
# Update the path below so that search_space.json is written to the same directory as your main.py file
# Note that your source code is cloned to /mnt/data/src
- name: search-space
path: /mnt/data/src/workflows/hyperparameter-tuning/mnist/search_space.json
raw:
data: '{{workflow.parameters.search-space}}'
outputs:
artifacts:
- name: output
path: /mnt/output
optional: true
container:
image: onepanel/dl:0.17.0
args:
- --config
# [CHANGE] Update the path below to point to config.yaml path as described above
- /mnt/data/src/workflows/hyperparameter-tuning/mnist/config.yaml
workingDir: /mnt
volumeMounts:
- name: hyperparamtuning-data
mountPath: /mnt/data
- name: hyperparamtuning-output
mountPath: /mnt/output
nodeSelector:
{{.NodePoolLabel}}: '{{workflow.parameters.sys-node-pool}}'
sidecars:
- name: nni-web-ui
image: onepanel/nni-web-ui:0.17.0
env:
- name: ONEPANEL_INTERACTIVE_SIDECAR
value: 'true'
ports:
- containerPort: 9000
name: nni
- name: tensorboard
image: onepanel/dl:0.17.0
command:
- sh
- '-c'
env:
- name: ONEPANEL_INTERACTIVE_SIDECAR
value: 'true'
args:
# Read logs from /mnt/output/tensorboard - /mnt/output is auto-mounted from volumeMounts
- tensorboard --logdir /mnt/output/tensorboard
ports:
- containerPort: 6006
name: tensorboard
# Use the metrics-writer tasks to write best metrics to Workflow
- name: metrics-writer
inputs:
artifacts:
- name: sys-metrics
path: /tmp/sys-metrics.json
- git:
repo: https://github.com/onepanelio/templates.git
revision: v0.18.0
name: src
path: /mnt/src
container:
image: onepanel/python-sdk:v0.16.0
command:
- python
- -u
args:
- /mnt/src/tasks/metrics-writer/main.py
- --from_file=/tmp/sys-metrics.json
# [CHANGE] Volumes that will mount to /mnt/data (annotated data) and /mnt/output (models, checkpoints, logs)
# Update this depending on your annotation data, model, checkpoint, logs, etc. sizes
# Example values: 250Mi, 500Gi, 1Ti
volumeClaimTemplates:
- metadata:
name: hyperparamtuning-data
@@ -64,120 +187,3 @@ volumeClaimTemplates:
resources:
requests:
storage: 20Gi
templates:
- name: main
dag:
tasks:
- name: hyperparameter-tuning
template: hyperparameter-tuning
- name: workflow-metrics-writer
template: workflow-metrics-writer
dependencies: [hyperparameter-tuning]
arguments:
# Use sys-metrics artifact output from hyperparameter-tuning Task
artifacts:
- name: best-metrics
from: "{{tasks.hyperparameter-tuning.outputs.artifacts.sys-metrics}}"
- name: hyperparameter-tuning
inputs:
artifacts:
- name: src
git:
repo: '{{workflow.parameters.source}}'
revision: '{{workflow.parameters.revision}}'
path: /mnt/data/src
- name: config
path: /mnt/data/src/workflows/hyperparameter-tuning/mnist/config.yaml
raw:
data: '{{workflow.parameters.config}}'
- name: search-space
path: /mnt/data/src/workflows/hyperparameter-tuning/mnist/search_space.json
raw:
data: '{{workflow.parameters.search-space}}'
outputs:
artifacts:
- name: output
path: /mnt/output
optional: true
container:
image: onepanel/dl:0.17.0
args:
- --config
- /mnt/data/src/workflows/hyperparameter-tuning/mnist/config.yaml
workingDir: /mnt
volumeMounts:
- name: hyperparamtuning-data
mountPath: /mnt/data
- name: hyperparamtuning-output
mountPath: /mnt/output
nodeSelector:
{{.NodePoolLabel}}: '{{workflow.parameters.sys-node-pool}}'
sidecars:
- name: nni-web-ui
image: 'onepanel/nni-web-ui:0.17.0'
env:
- name: ONEPANEL_INTERACTIVE_SIDECAR
value: 'true'
ports:
- containerPort: 9000
name: nni
- name: tensorboard
image: 'tensorflow/tensorflow:2.3.0'
command:
- sh
- '-c'
env:
- name: ONEPANEL_INTERACTIVE_SIDECAR
value: 'true'
args:
# Read logs from /mnt/output/tensorboard - /mnt/output is auto-mounted from volumeMounts
- tensorboard --logdir /mnt/output/tensorboard
ports:
- containerPort: 6006
name: tensorboard
- name: workflow-metrics-writer
inputs:
artifacts:
- name: best-metrics
path: /tmp/sys-metrics.json
script:
image: onepanel/python-sdk:v0.16.0
command: [python, '-u']
source: |
import os
import json
import onepanel.core.api
from onepanel.core.api.models.metric import Metric
from onepanel.core.api.rest import ApiException
from onepanel.core.api.models import Parameter
# Load Task A metrics
with open('/tmp/sys-metrics.json') as f:
metrics = json.load(f)
with open('/var/run/secrets/kubernetes.io/serviceaccount/token') as f:
token = f.read()
# Configure API authorization
configuration = onepanel.core.api.Configuration(
host = os.getenv('ONEPANEL_API_URL'),
api_key = {
'authorization': token
}
)
configuration.api_key_prefix['authorization'] = 'Bearer'
# Call SDK method to save metrics
with onepanel.core.api.ApiClient(configuration) as api_client:
api_instance = onepanel.core.api.WorkflowServiceApi(api_client)
namespace = '{{workflow.namespace}}'
uid = '{{workflow.name}}'
body = onepanel.core.api.AddWorkflowExecutionsMetricsRequest()
body.metrics = metrics
try:
api_response = api_instance.add_workflow_execution_metrics(namespace, uid, body)
print('Metrics added.')
except ApiException as e:
print("Exception when calling WorkflowServiceApi->add_workflow_execution_metrics: %s\n" % e)