Dockerize TG Freq

.github/workflows/pipeline-select-galaxy.yaml

@@ -79,10 +79,14 @@ jobs:
     uses: ./.github/workflows/tg-unit-tests-impl.yaml
     if: ${{ inputs.tg-unit }}
   tg-frequent-tests:
+    if: ${{ inputs.tg-frequent }}
     needs: build-artifact
     secrets: inherit
     uses: ./.github/workflows/tg-frequent-tests-impl.yaml
-    if: ${{ inputs.tg-frequent }}
+    with:
+      docker-image: ${{ needs.build-artifact.outputs.ci-build-docker-image }}
+      wheel-artifact-name: ${{ needs.build-artifact.outputs.wheel-artifact-name }}
+      build-artifact-name: ${{ needs.build-artifact.outputs.build-artifact-name }}
   tg-model-perf-tests:
     needs: build-artifact
     secrets: inherit

.github/workflows/tg-frequent-tests-impl.yaml

@@ -2,6 +2,16 @@ name: "[internal] TG frequent tests"
 
 on:
   workflow_call:
+    inputs:
+      docker-image:
+        required: true
+        type: string
+      wheel-artifact-name:
+        required: true
+        type: string
+      build-artifact-name:
+        required: true
+        type: string
 
 jobs:
   tg-frequent-tests:
@@ -13,36 +23,62 @@ jobs:
           { name: "TG resnet50 frequent tests", arch: wormhole_b0, model: resnet50, timeout: 90, owner_id: U013121KDH9}, # Austin Ho
           { name: "TG unit/distributed frequent tests", arch: wormhole_b0, model: unit, timeout: 90, owner_id: XXXXX}, # Add owner
         ]
-    env:
-      ARCH_NAME: ${{ matrix.test-group.arch }}
-      LOGURU_LEVEL: INFO
-      LD_LIBRARY_PATH: ${{ github.workspace }}/build/lib
     runs-on:
       - arch-wormhole_b0
       - config-tg
       - in-service
       - bare-metal
       - pipeline-functional
+    container:
+      image: ${{ inputs.docker-image }}
+      env:
+        TT_METAL_HOME: /work
+        PYTHONPATH: /work
+        LD_LIBRARY_PATH: /work/build/lib
+        LOGURU_LEVEL: INFO
+        ARCH_NAME: ${{ matrix.test-group.arch }}
+      volumes:
+        - ${{ github.workspace }}/docker-job:/work # Subdir to workaround https://github.com/actions/runner/issues/691
+        - /dev/hugepages-1G:/dev/hugepages-1G
+        - /mnt/MLPerf:/mnt/MLPerf
+      options: "--device /dev/tenstorrent"
+    defaults:
+      run:
+        shell: bash
+        working-directory: /work # https://github.com/actions/runner/issues/878
     steps:
-      - uses: tenstorrent/tt-metal/.github/actions/checkout-with-submodule-lfs@main
-      - name: Set up dynamic env vars for build
-        run: |
-          echo "TT_METAL_HOME=$(pwd)" >> $GITHUB_ENV
+      - name: ⬇️ Checkout
+        uses: actions/checkout@v4
+        with:
+          submodules: recursive
+          path: docker-job # Here be dragons; keep it scoped to our desired volume, yet must be under github.workspace and be sure to clean up at the end
       - uses: actions/download-artifact@v4
         with:
-          name: TTMetal_build_any
+          name: ${{ inputs.build-artifact-name }}
+          path: /work
       - name: Extract files
         run: tar -xvf ttm_any.tar
-      - uses: ./.github/actions/install-python-deps
+      - name: ⬇️ Download Wheel
+        uses: actions/download-artifact@v4
+        with:
+          name: ${{ inputs.wheel-artifact-name }}
+          path: /work
+      - name: Install Wheel
+        run: |
+          WHEEL_FILENAME=$(ls -1 *.whl)
+          pip3 install $WHEEL_FILENAME
       - name: Run frequent regression tests
         timeout-minutes: ${{ matrix.test-group.timeout }}
         run: |
-          source ${{ github.workspace }}/python_env/bin/activate
-          cd $TT_METAL_HOME
-          export PYTHONPATH=$TT_METAL_HOME
           ./tests/scripts/run_tests.sh --tt-arch wormhole_b0 --pipeline-type frequent_tg_device --dispatch-mode "" --model ${{ matrix.test-group.model }}
       - uses: ./.github/actions/slack-report
         if: ${{ failure() }}
         with:
           slack_webhook_url: ${{ secrets.SLACK_WEBHOOK_URL }}
           owner: ${{ matrix.test-group.owner_id }}
+      - name: Cleanup
+        if: always()
+        run: |
+          # We are forced to checkout the repo into a subdir of the host's workdir; this pollutes the host
+          # with root-owned files.  Be sure to clean up after ourselves in case we're on a non-ephemeral runner.
+          rm -rf /__w/tt-metal/tt-metal/docker-job

.github/workflows/tg-frequent-tests.yaml

@@ -9,7 +9,13 @@ jobs:
   build-artifact:
     uses: ./.github/workflows/build-artifact.yaml
     secrets: inherit
+    with:
+      build-wheel: true
   tg-frequent-tests:
     needs: build-artifact
     secrets: inherit
     uses: ./.github/workflows/tg-frequent-tests-impl.yaml
+    with:
+      docker-image: ${{ needs.build-artifact.outputs.ci-build-docker-image }}
+      wheel-artifact-name: ${{ needs.build-artifact.outputs.wheel-artifact-name }}
+      build-artifact-name: ${{ needs.build-artifact.outputs.build-artifact-name }}