Merge branch 'add_some_jobs_dracorno' into 'main'

[READY]Add jobs 'hps_plugin_benchmark','147gb_model_benchmark'

See merge request dl/hugectr/hugectr!1435

ci/dracorno/ci.yml

@@ -151,7 +151,7 @@ py_single_node:
   variables:
     GPFSFOLDER: $DRACO_LOGDIR/py_single_node
     CONT: $TRAIN_IMAGE_VERSIONED
-    MOUNTS: ${DRACO_DATASET}:${DATASET_MOUNT},${DRACO_WDL_PARQUET_DATASET}:${NEW_CRITEO_MOUNT}
+    MOUNTS: ${DRACO_DATASET}:${DATASET_MOUNT},${DRACO_DATASET_NEW_CRITEO}:${NEW_CRITEO_MOUNT}
     TEST_CMD: ./ci/integration_test/py_interface/py_single_node.sub
 
 hugectr2onnx:
@@ -175,7 +175,7 @@ ebc_multi_node:
     GPFSFOLDER: $DRACO_LOGDIR/ebc_multi_node
     CONT: $TRAIN_IMAGE_MULTINODE_VERSIONED
     MOUNTS: ${DRACO_DATASET_NEW_CRITEO}:${DATASET_MOUNT}
-    WALLTIME: "00:45:00"
+    WALLTIME: "01:00:00"
     DGXNNODES: 2
     TEST_CMD: ./ci/integration_test/ebc/ebc.sub
 
@@ -214,6 +214,30 @@ s3_backend_test:
     DGXNNODES: 1
     TEST_CMD: ./ci/integration_test/s3/s3_backend_test.sub
 
+hps_plugin_benchmark:
+  extends: .dracorno_test_job
+  needs:
+    - pipeline: $PARENT_PIPELINE_ID
+      job: build_tf_hps_trt_plugin
+  variables:
+    GPFSFOLDER: $DRACO_LOGDIR/hps_plugin_benchmark
+    CONT: $TF_TRT_IMAGE_VERSIONED
+    MOUNTS: ${DATA_PREFIX}/fs1/projects/gpu_compute/datasets/hugectr-ci/hpc-hugectr/hps_tf_benchmark/hps_plugin_ci_model_repo:/model_repo,${DATA_PREFIX}/fs1/projects/gpu_compute/datasets/hugectr-ci/hpc-hugectr/hps_tf_benchmark/perf_data:/perf_data
+    WALLTIME: "00:45:00"
+    TEST_CMD: ./ci/benchmark/hps_plugin_benchmark/run.sub
+
+147gb_model_benchmark:
+  extends: .dracorno_test_job
+  needs:
+    - pipeline: $PARENT_PIPELINE_ID
+      job: build_tf_hps_trt_plugin
+  variables:
+    GPFSFOLDER: $DRACO_LOGDIR/147gb_model_benchmark
+    CONT: $TF_TRT_IMAGE_VERSIONED
+    MOUNTS: ${DATA_PREFIX}/fs1/projects/gpu_compute/datasets/hugectr-ci/hpc-hugectr/hps_tf_benchmark/147gb_ci_model_repo:/model_repo
+    WALLTIME: "00:45:00"
+    TEST_CMD: ./ci/benchmark/147gb_model_benchmark/run.sub
+
 #SOK ut tests
 sparse_operation_kit_ut-TF1:
   extends:

ci/template.yml

@@ -384,6 +384,14 @@ stages:
     GIT_CLONE_PATH: /lustre/fsw/devtech/hpc-hugectr/hugectr-ci/$CI_CONCURRENT_ID/$CI_PROJECT_NAME
   stage: post_test
 
+.dracorno_post_test_job:
+  extends:
+    - .dracorno_test_job
+    - .hugectr:rules:test_in_child
+  variables:
+    WALLTIME: "00:30:00"
+  stage: post_test
+
 .cluster_post_test_job_daily:
   extends:
     - .cluster_test_job