feat: add pr e2e test (#822)

.github/workflows/pr-e2e-test.yml

@@ -12,19 +12,36 @@ jobs:
     runs-on: self-hosted
     env:
       CUDA_VISIBLE_DEVICES: 6
+
     steps:
-    - uses: actions/checkout@v2
+    - name: Checkout code
-    - name: Check GPU
+      uses: actions/checkout@v3
+
+    - name: Install dependencies
       run: |
-        if ! command -v nvidia-smi &> /dev/null; then
+        pip install --upgrade pip
-          echo "nvidia-smi not found. Is CUDA installed?"
+        pip install -e "python[all]"
-          exit 1
+        pip install flashinfer -i https://flashinfer.ai/whl/cu121/torch2.3/ --force-reinstall
-        fi
+        pip install --upgrade transformers
-        nvidia-smi || exit 1
+
-    - name: Environment Info
+    - name: Launch server and run benchmark
       run: |
-        echo "Working directory: $(pwd)"
+        python3 -m sglang.launch_server --model /home/lmzheng/zhyncs/Meta-Llama-3.1-8B-Instruct --port 8413 &
-        echo "CUDA_VISIBLE_DEVICES: $CUDA_VISIBLE_DEVICES"
+
-    - name: Run Tests
+        echo "Waiting for server to start..."
-      run: |
+        for i in {1..60}; do
-        echo "Running tests..."
+          if curl -s http://127.0.0.1:8413/health; then
+            echo "Server is up!"
+            break
+          fi
+          if [ $i -eq 60 ]; then
+            echo "Server failed to start within 60 seconds"
+            exit 1
+          fi
+          sleep 1
+        done
+
+        python3 -m sglang.bench_serving --backend sglang --port 8413
+
+        echo "Stopping server..."
+        kill -9 $(ps aux | grep sglang | grep Meta-Llama-3.1-8B-Instruct | grep -v grep | awk '{print $2}')