Merge branch 'users/stbaione/sglang-benchmark-fix' of https://github.com/nod-ai/shark-ai into fix-sglang-benchmark-missing-models

Author: renxida

.github/workflows/ci-sglang-benchmark.yml

@@ -20,6 +20,8 @@ name: SGLang Llama Benchmarking Tests
 
 on:
   workflow_dispatch:
+  # TODO: Remove after CI validation
+  pull_request:
   schedule:
     # Weekdays at 11:00 AM UTC = 03:00 AM PST / 04:00 AM PDT
     - cron: "0 11 * * 1-5"
@@ -80,6 +82,9 @@ jobs:
 
           pip freeze
 
+      - name: Login to huggingface
+        run: huggingface-cli login --token ${{ secrets.HF_TOKEN }}
+
       - name: Run Shortfin Benchmark Tests
         run: |
           source ${VENV_DIR}/bin/activate
@@ -110,11 +115,25 @@ jobs:
         with:
           python-version: ${{matrix.version}}
 
-      - name: Install SGLang
+      - name: Install deps
         run: |
           python -m pip install --no-compile --upgrade pip
+
           pip install "git+https://github.com/nod-ai/sglang.git#subdirectory=python"
 
+          pip install pytest
+
+          # Use newest possible releases to be able to track commits that may
+          # cause errors or performance changes.
+          pip install -r requirements-iree-unpinned.txt
+
+          pip install --no-compile \
+            -r sharktank/requirements-tests.txt \
+            -r shortfin/requirements-tests.txt \
+            -e sharktank/ shortfin/
+
+          pip freeze
+
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
 
@@ -158,19 +177,26 @@ jobs:
         run: |
           pytest -v app_tests/benchmark_tests/llm/sglang_benchmarks/sglang_benchmark_test.py --port 30000 --log-cli-level=INFO --html=sglang_index.html --self-contained-html
 
+      - name: Upload pytest report
+        uses: actions/upload-artifact@65c4c4a1ddee5b72f698fdd19549f0f0fb45cf08
+        with:
+          name: sglang_benchmark
+          path: sglang_index.html
+
+  # Ensure that the container is always cleaned up after job
+  container_cleanup:
+    needs: benchmark_sglang
+    name: "Docker Cleanup"
+    if: always()
+    runs-on: mi300x-3
+    steps:
       - name: Stop sglang-server
         run: docker stop sglang-server || true # Stop container if it's running
 
       # Deleting image after run due to large disk space requirement (83 GB)
       - name: Cleanup SGLang Image
         run: docker image rm lmsysorg/sglang:v0.3.5.post1-rocm620
 
-      - name: Upload pytest report
-        uses: actions/upload-artifact@65c4c4a1ddee5b72f698fdd19549f0f0fb45cf08
-        with:
-          name: sglang_benchmark
-          path: sglang_index.html
-
   merge_and_upload_reports:
     name: "Merge and upload benchmark reports"
     needs: [benchmark_shortfin, benchmark_sglang]