test lcnc on g6

deepjavalibrary · Apr 5, 2024 · b834f78 · b834f78
1 parent 276cd84
commit b834f78
Showing 1 changed file with 38 additions and 4 deletions.
diff --git a/.github/workflows/lmi-no-code.yml b/.github/workflows/lmi-no-code.yml
@@ -32,6 +32,24 @@ jobs:
           --fail \
           | jq '.token' | tr -d '"' )
           ./start_instance.sh action_g5 $token djl-serving
+      - name: Create new G6.12xl instance
+        id: create_gpu_g612_1
+        run: |
+          cd /home/ubuntu/djl_benchmark_script/scripts
+          token=$( curl -X POST -H "Authorization: token ${{ secrets.ACTION_RUNNER_PERSONAL_TOKEN }}" \
+          https://api.github.com/repos/deepjavalibrary/djl-serving/actions/runners/registration-token \
+          --fail \
+          | jq '.token' | tr -d '"' )
+          ./start_instance.sh action_g6 $token djl-serving
+      - name: Create new G6.12xl instance
+        id: create_gpu_g612_2
+        run: |
+          cd /home/ubuntu/djl_benchmark_script/scripts
+          token=$( curl -X POST -H "Authorization: token ${{ secrets.ACTION_RUNNER_PERSONAL_TOKEN }}" \
+          https://api.github.com/repos/deepjavalibrary/djl-serving/actions/runners/registration-token \
+          --fail \
+          | jq '.token' | tr -d '"' )
+          ./start_instance.sh action_g6 $token djl-serving
       - name: Create new P4d instance
         id: create_gpu_p4d
         run: |
@@ -44,6 +62,8 @@ jobs:
     outputs:
       g512_instance_id_1: ${{ steps.create_gpu_g512_1.outputs.action_g5_instance_id }}
       g512_instance_id_2: ${{ steps.create_gpu_g512_2.outputs.action_g5_instance_id }}
+      g612_instance_id_1: ${{ steps.create_gpu_g612_1.outputs.action_g6_instance_id }}
+      g612_instance_id_2: ${{ steps.create_gpu_g612_2.outputs.action_g6_instance_id }}
       p4d_instance_id: ${{ steps.create_gpu_p4d.outputs.action_lmic_p4d_instance_id }}
 
   p4d-no-code-tests:
@@ -139,14 +159,15 @@ jobs:
           name: no-code-p4d-${{ matrix.container }}-logs
           path: tests/integration/logs/
 
-  g512-no-code-tests:
-    runs-on: [self-hosted, g5]
+  g-series-no-code-tests:
+    runs-on: [self-hosted, "${{ matrix.machine }}"]
     timeout-minutes: 240
     needs: create-runners
     strategy:
       fail-fast: false
       matrix:
         container: [tensorrt-llm, deepspeed]
+        machine: [g5, g6]
     steps:
       - uses: actions/checkout@v4
       - name: Clean env
@@ -279,17 +300,30 @@ jobs:
           path: tests/integration/logs/
 
 
-  stop-runners:
+  stop-runners-gseries:
     if: always()
     runs-on: [self-hosted, scheduler]
-    needs: [create-runners, g512-no-code-tests, p4d-no-code-tests]
+    needs: [create-runners, g-series-no-code-tests]
     steps:
       - name: Stop all instances
         run: |
           cd /home/ubuntu/djl_benchmark_script/scripts
           instance_id=${{ needs.create-runners.outputs.g512_instance_id_1 }}
           ./stop_instance.sh $instance_id
           instance_id=${{ needs.create-runners.outputs.g512_instance_id_2 }}
+          cd /home/ubuntu/djl_benchmark_script/scripts
+          instance_id=${{ needs.create-runners.outputs.g612_instance_id_1 }}
+          ./stop_instance.sh $instance_id
+          instance_id=${{ needs.create-runners.outputs.g612_instance_id_2 }}
           ./stop_instance.sh $instance_id
+
+  stop-runners-p4d:
+    if: always()
+    runs-on: [self-hosted, scheduler]
+    needs: [create-runners, p4d-no-code-tests]
+    steps:
+      - name: Stop all instances
+        run: |
+          cd /home/ubuntu/djl_benchmark_script/scripts
           instance_id=${{ needs.create-runners.outputs.p4d_instance_id }}
           ./stop_instance.sh $instance_id