Merge branch 'main' into fix/deffered-imports

Author: mohamedzeidan2021

.github/workflows/codebuild-ci.yml

@@ -2,8 +2,7 @@ name: PR Checks
 on:
   pull_request_target:
       branches:
-          - "master*"
-          - "main*"
+          - "*"
 
 concurrency:
     group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.head_ref }}

CHANGELOG.md

@@ -1,5 +1,11 @@
 # Changelog
 
+## v3.1.0 (2025-08-13)
+
+### Features
+
+ * Task Governance feature for training jobs.
+
 ## v3.0.2 (2025-07-31)
 
 ### Features

helm_chart/HyperPodHelmChart/charts/health-monitoring-agent/templates/health-monitoring-agent.yaml

@@ -111,6 +111,7 @@ spec:
                       - ml.g6e.48xlarge
                       - ml.trn2.48xlarge
                       - ml.p6-b200.48xlarge
+                      - ml.p6e-gb200.36xlarge
       containers:
         - name: health-monitoring-agent
           args:

helm_chart/HyperPodHelmChart/values.yaml

@@ -180,6 +180,8 @@ nvidia-device-plugin:
               - ml.p5.48xlarge
               - ml.p5e.48xlarge
               - ml.p5en.48xlarge
+              - ml.p6-b200.48xlarge
+              - ml.p6e-gb200.36xlarge
   tolerations:
     - key: nvidia.com/gpu
       operator: Exists
@@ -197,6 +199,7 @@ aws-efa-k8s-device-plugin:
   devicePlugin:
     enabled: true
   supportedInstanceLabels:
+    # https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/efa.html#efa-instance-types
     values:
       - ml.c5n.9xlarge
       - ml.c5n.18xlarge
@@ -237,6 +240,8 @@ aws-efa-k8s-device-plugin:
       - ml.p5.48xlarge
       - ml.p5e.48xlarge
       - ml.p5en.48xlarge
+      - ml.p6-b200.48xlarge
+      - ml.p6e-gb200.36xlarge
       - ml.r7i.large
       - ml.r7i.xlarge
       - ml.r7i.2xlarge

hyperpod-custom-inference-template/pyproject.toml

@@ -20,4 +20,5 @@ include-package-data = true
 
 [tool.setuptools.package-data]
 # for each versioned subpackage, include schema.json
-"hyperpod_custom_inference_template.v1_0" = ["schema.json"]
+"*" = ["schema.json"]
+

hyperpod-jumpstart-inference-template/pyproject.toml

@@ -20,4 +20,5 @@ include-package-data = true
 
 [tool.setuptools.package-data]
 # for each versioned subpackage, include schema.json
-"hyperpod_jumpstart_inference_template.v1_0" = ["schema.json"]
+"*" = ["schema.json"]
+

hyperpod-pytorch-job-template/CHANGELOG.md

@@ -1,3 +1,9 @@
+## v1.1.0 (2025-08-14)
+
+### Features
+
+ * Added parameters for task governance feature
+
 ## v1.0.2 (2025-07-31)
 
 ### Features

hyperpod-pytorch-job-template/hyperpod_pytorch_job_template/registry.py

@@ -10,11 +10,13 @@
 # distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF
 # ANY KIND, either express or implied. See the License for the specific
 # language governing permissions and limitations under the License.
-from .v1_0.model import PyTorchJobConfig  # Import your model
+from .v1_0 import model as v1_0_model # Import your model
+from .v1_1 import model as v1_1_model
 from typing import Dict, Type
 from pydantic import BaseModel
 
 # Direct version-to-model mapping
 SCHEMA_REGISTRY: Dict[str, Type[BaseModel]] = {
-    "1.0": PyTorchJobConfig,
+    "1.0": v1_0_model.PyTorchJobConfig,
+    "1.1": v1_1_model.PyTorchJobConfig,
 }

hyperpod-pytorch-job-template/hyperpod_pytorch_job_template/v1_0/model.py

@@ -353,6 +353,7 @@ def to_domain(self) -> Dict:
         result = {
             "name": self.job_name,
             "namespace": self.namespace,
+            "labels": metadata_labels,
             "spec": job_kwargs,
         }
         return result

hyperpod-pytorch-job-template/hyperpod_pytorch_job_template/v1_1/__init__.py

@@ -0,0 +1,7 @@
+from .model import PyTorchJobConfig
+
+def validate(data: dict):
+    return PyTorchJobConfig(**data)
+
+
+__all__ = ["validate", "PyTorchJobConfig"]