Adding mlscale_onduty tag for dags that are actively monitored by the…

… ML@Scale onduty (#501)
GoogleCloudPlatform · Dec 7, 2024 · 846a04a · 846a04a
1 parent 4afadb4
commit 846a04a
Show file tree

Hide file tree

Showing 12 changed files with 19 additions and 9 deletions.
diff --git a/dags/multipod/jax_functional_tests.py b/dags/multipod/jax_functional_tests.py
@@ -27,7 +27,7 @@
 with models.DAG(
     dag_id="jax_functional_tests",
     schedule=SCHEDULED_TIME,
-    tags=["multipod_team", "jax"],
+    tags=["multipod_team", "jax", "mlscale_onduty"],
     start_date=datetime.datetime(2024, 10, 23),
     catchup=False,
 ) as dag:

diff --git a/dags/multipod/legacy.py b/dags/multipod/legacy.py
@@ -48,7 +48,14 @@
 with models.DAG(
     dag_id=f"multipod_legacy_xlml",
     schedule=SCHEDULED_TIME,
-    tags=["multipod_team", "xlml", "legacy", "stable", "nightly"],
+    tags=[
+        "multipod_team",
+        "xlml",
+        "legacy",
+        "stable",
+        "nightly",
+        "mlscale_onduty",
+    ],
     start_date=datetime.datetime(2024, 1, 10),
     catchup=False,
     concurrency=2,

diff --git a/dags/multipod/maxtext_configs_aot_hybridsim.py b/dags/multipod/maxtext_configs_aot_hybridsim.py
@@ -84,7 +84,7 @@ def hybridsim_compile_and_run(test_group_id):
 with models.DAG(
     dag_id="maxtext_configs_aot_hybridsim",
     schedule=SCHEDULED_TIME,
-    tags=["multipod_team", "maxtext", "nightly"],
+    tags=["multipod_team", "maxtext", "nightly", "mlscale_onduty"],
     start_date=datetime.datetime(2024, 2, 19),
     catchup=False,
     concurrency=10,

diff --git a/dags/multipod/maxtext_convergence.py b/dags/multipod/maxtext_convergence.py
@@ -29,7 +29,7 @@
 with models.DAG(
     dag_id="maxtext_convergence",
     schedule=SCHEDULED_TIME,
-    tags=["multipod_team", "maxtext", "stable"],
+    tags=["multipod_team", "maxtext", "stable", "mlscale_onduty"],
     start_date=datetime.datetime(2024, 3, 1),
     catchup=False,
     concurrency=2,

diff --git a/dags/multipod/maxtext_end_to_end.py b/dags/multipod/maxtext_end_to_end.py
@@ -31,7 +31,7 @@
 with models.DAG(
     dag_id="maxtext_end_to_end",
     schedule=SCHEDULED_TIME,
-    tags=["multipod_team", "maxtext", "stable", "nightly"],
+    tags=["multipod_team", "maxtext", "stable", "nightly", "mlscale_onduty"],
     start_date=datetime.datetime(2024, 1, 19),
     catchup=False,
 ) as dag:

diff --git a/dags/multipod/maxtext_gpu_end_to_end.py b/dags/multipod/maxtext_gpu_end_to_end.py
@@ -30,7 +30,7 @@
 with models.DAG(
     dag_id="maxtext_gpu_end_to_end",
     schedule=SCHEDULED_TIME,
-    tags=["multipod_team", "maxtext", "stable", "nightly"],
+    tags=["multipod_team", "maxtext", "stable", "nightly", "mlscale_onduty"],
     start_date=datetime.datetime(2024, 1, 19),
     catchup=False,
 ) as dag:

diff --git a/dags/multipod/maxtext_v5e_configs_perf.py b/dags/multipod/maxtext_v5e_configs_perf.py
@@ -46,7 +46,7 @@
 with models.DAG(
     dag_id="maxtext_v5e_configs_perf",
     schedule=SCHEDULED_TIME,
-    tags=["multipod_team", "maxtext", "stable", "nightly"],
+    tags=["multipod_team", "maxtext", "stable", "nightly", "mlscale_onduty"],
     start_date=datetime.datetime(2024, 2, 19),
     catchup=False,
 ) as dag:

diff --git a/dags/multipod/mxla_maxtext_nightly.py b/dags/multipod/mxla_maxtext_nightly.py
@@ -28,7 +28,7 @@
 with models.DAG(
     dag_id="mxla_maxtext_nightly",
     schedule=SCHEDULED_TIME,
-    tags=["multipod_team", "maxtext", "nightly"],
+    tags=["multipod_team", "maxtext", "nightly", "mlscale_onduty"],
     start_date=datetime.datetime(2024, 1, 10),
     catchup=False,
 ) as dag:

diff --git a/dags/multipod/pytorch.py b/dags/multipod/pytorch.py
@@ -28,7 +28,7 @@
 with models.DAG(
     dag_id="pytorch_multislice",
     schedule=SCHEDULED_TIME,
-    tags=["multipod_team", "pytorch", "nightly"],
+    tags=["multipod_team", "pytorch", "nightly", "mlscale_onduty"],
     start_date=datetime.datetime(2024, 3, 1),
     catchup=False,
     concurrency=2,

diff --git a/dags/sparsity_diffusion_devx/maxtext_moe_tpu_e2e.py b/dags/sparsity_diffusion_devx/maxtext_moe_tpu_e2e.py
@@ -38,6 +38,7 @@
         "tpu",
         "stable",
         "nightly",
+        "mlscale_onduty",
     ],
     start_date=datetime.datetime(2024, 11, 14),
     catchup=False,

diff --git a/dags/sparsity_diffusion_devx/project_bite_gpu_e2e.py b/dags/sparsity_diffusion_devx/project_bite_gpu_e2e.py
@@ -35,6 +35,7 @@
         "gpu",
         "axlearn",
         "bite",
+        "mlscale_onduty",
     ],
     start_date=datetime.datetime(2024, 11, 12),
     catchup=False,

diff --git a/dags/sparsity_diffusion_devx/project_bite_tpu_e2e.py b/dags/sparsity_diffusion_devx/project_bite_tpu_e2e.py
@@ -34,6 +34,7 @@
         "tpu",
         "axlearn",
         "bite",
+        "mlscale_onduty",
     ],
     start_date=datetime.datetime(2024, 4, 4),
     catchup=False,