Add the reworked deployment code with additional flags

determined-ai · Mar 23, 2024 · 94737e6 · 94737e6
1 parent 1099bfa
commit 94737e6
Show file tree

Hide file tree

Showing 4 changed files with 8 additions and 9 deletions.
diff --git a/examples/3d-brain-mri/container/deploy/common.py b/examples/3d-brain-mri/container/deploy/common.py
@@ -122,19 +122,19 @@ def parse_args():
     parser.add_argument(
         "--tolerations",
         type=csv_,
-        help="a comma separated list of tolerations to apply to the deployment in the format of key=value",
+        help="A comma separated list of tolerations to apply to the deployment in the format of key=value",
         default=None,
     )
     parser.add_argument(
         "--resource-requests",
         type=csv_,
-        help="",
+        help="A comma separated list of resource requests to apply to the deployment in the format of key=value",
         default=None,
     )
     parser.add_argument(
         "--resource-limits",
         type=csv_,
-        help="",
+        help="A comma separated list of resource limits to apply to the deployment in the format of key=value",
         default=None,
     )
     return parser.parse_args()

diff --git a/examples/3d-brain-mri/experiment/const.yaml b/examples/3d-brain-mri/experiment/const.yaml
@@ -46,5 +46,4 @@ entrypoint: model_code.model_def:MRIVnetTrial
 profiling:
   enabled: True
 resources:
-  slots_per_trial: 4
-  # resource_pool: A100
+  slots_per_trial: 4
diff --git a/examples/3d-brain-mri/pipelines/_on_prem_deployment-pipeline.json b/examples/3d-brain-mri/pipelines/_on_prem_deployment-pipeline.json
@@ -17,9 +17,9 @@
       "/bin/sh"
     ],
     "stdin": [
-      "python deploy.py --deployment-name pdk-3d-brain-mri-deploy --service-account-name pach-deploy --tolerations accelerator=Tesla-T4 -resource-requests cpu=10,memory=8Gi --resource-limits cpu=10,memory=8Gi"
+      "python deploy.py --deployment-name pdk-3d-brain-mri-deploy --service-account-name pach-deploy --max-request-size 400553500 --max-response-size 400553500 --resource-requests nvidia.com/gpu=1,memory=8Gi --resource-limits nvidia.com/gpu=1,memory=8Gi --tolerations accelerator=Tesla-T4"
     ],
-    "image": "alemor/pdk:3d-brain-deploy-v0.0.3",
+    "image": "alemor/pdk:3d-brain-deploy-v0.0.5",
     "secrets": [
       {
         "name": "pipeline-secret",

diff --git a/examples/3d-brain-mri/pipelines/deployment-pipeline.json b/examples/3d-brain-mri/pipelines/deployment-pipeline.json
@@ -16,9 +16,9 @@
       "/bin/sh"
     ],
     "stdin": [
-      "python deploy.py --deployment-name pdk-3d-brain-mri-deploy --cloud-model-host gcp --cloud-model-bucket alemor-pdk-gcp-repo-models --resource-requests cpu=10,memory=8Gi --resource-limits cpu=20,memory=8Gi"
+      "python deploy.py --deployment-name pdk-3d-brain-mri-deploy --cloud-model-host gcp --cloud-model-bucket alemor-pdk-gcp-repo-models --git-url https://git@github.com:/determined-ai/pdk.git  --handler brain_mri_handler.py --git-ref dev_3dmri --sub-dir examples/3d-brain-mri/container/deploy --max-request-size 400553500 --max-response-size 400553500 --resource-requests nvidia.com/gpu=1,memory=8Gi --resource-limits nvidia.com/gpu=4,memory=8Gi --tolerations nvidia.com/gpu=present"
     ],
-    "image": "alemor/pdk:3d-brain-deploy-v0.0.4",
+    "image": "alemor/pdk:3d-brain-deploy-v0.0.5",
     "secrets": [
       {
         "name": "pipeline-secret",