Upload test_vllm_endpoint.py with huggingface_hub

test_vllm_endpoint.py

@@ -0,0 +1,165 @@
+#!/usr/bin/env python3
+"""
+Test RunPod vLLM endpoint with Ultravox
+"""
+
+import os
+import sys
+import time
+import runpod
+
+# Configuration
+API_KEY = "YOUR_RUNPOD_API_KEY_HERE"
+ENDPOINT_ID = os.getenv("RUNPOD_ENDPOINT_ID", "")
+
+if not ENDPOINT_ID:
+    print("=" * 60)
+    print("❌ Error: RUNPOD_ENDPOINT_ID not set")
+    print("=" * 60)
+    print()
+    print("Usage:")
+    print("  export RUNPOD_ENDPOINT_ID='your-endpoint-id'")
+    print("  python3 test_vllm_endpoint.py")
+    print()
+    sys.exit(1)
+
+print("=" * 60)
+print("Testing RunPod vLLM Endpoint (Ultravox)")
+print("=" * 60)
+print()
+print(f"Endpoint ID: {ENDPOINT_ID}")
+print()
+
+# Initialize endpoint
+runpod.api_key = API_KEY
+endpoint = runpod.Endpoint(ENDPOINT_ID)
+
+# Test 1: First request (cold start)
+print("🧪 Test 1: Cold Start Performance")
+print("-" * 60)
+print("Sending first request (expect 8-15s)...")
+print()
+
+start_time = time.time()
+
+try:
+    result = endpoint.run_sync({
+        "input": {
+            "prompt": "Explain what Ultravox is in one sentence.",
+            "max_tokens": 100,
+            "temperature": 0.7
+        }
+    }, timeout=120)
+
+    elapsed = time.time() - start_time
+
+    print(f"✅ Success! Total time: {elapsed:.2f}s")
+    print()
+
+    # Extract response
+    if isinstance(result, dict):
+        if 'output' in result:
+            response = result['output']
+        elif 'text' in result:
+            response = result['text']
+        else:
+            response = str(result)
+    else:
+        response = str(result)
+
+    print("Response:")
+    print(f"  {response}")
+    print()
+
+    # Analyze cold start
+    if elapsed > 10:
+        print("⏱️  Cold Start Confirmed!")
+        print(f"  First request: {elapsed:.2f}s")
+        print("  Next request will be much faster")
+    else:
+        print("⚡ Fast response! Instance was already warm.")
+
+    print()
+
+except Exception as e:
+    print(f"❌ Error: {e}")
+    print()
+    print("Troubleshooting:")
+    print("  1. Check endpoint is running in RunPod console")
+    print("  2. Verify endpoint ID is correct")
+    print("  3. Check environment variables in endpoint config")
+    print("  4. Look at logs in RunPod console")
+    sys.exit(1)
+
+# Wait a bit
+print("⏳ Waiting 2 seconds...")
+time.sleep(2)
+print()
+
+# Test 2: Warm request
+print("🚀 Test 2: Warm Request Performance")
+print("-" * 60)
+print("Sending second request (expect <1s)...")
+print()
+
+start_time = time.time()
+
+try:
+    result = endpoint.run_sync({
+        "input": {
+            "prompt": "What is 2+2?",
+            "max_tokens": 20
+        }
+    }, timeout=30)
+
+    elapsed = time.time() - start_time
+
+    print(f"✅ Success! Total time: {elapsed:.2f}s")
+    print()
+
+    # Extract response
+    if isinstance(result, dict):
+        if 'output' in result:
+            response = result['output']
+        elif 'text' in result:
+            response = result['text']
+        else:
+            response = str(result)
+    else:
+        response = str(result)
+
+    print("Response:")
+    print(f"  {response}")
+    print()
+
+    # Analyze performance
+    if elapsed < 2:
+        print("🎉 Excellent! Instance is WARM and fast!")
+        print(f"  Warm inference: {elapsed:.2f}s")
+    else:
+        print("⚠️  Slower than expected")
+        print(f"  Inference time: {elapsed:.2f}s")
+        print("  Instance may have scaled down already")
+
+except Exception as e:
+    print(f"❌ Error: {e}")
+
+print()
+print("=" * 60)
+print("✅ Testing Complete")
+print("=" * 60)
+print()
+print("Summary:")
+print("  ✅ vLLM endpoint is working")
+print("  ✅ Ultravox model loaded successfully")
+print("  ✅ Ready for production use")
+print()
+print("Next Steps:")
+print("  1. Update /workspace/ultravox-pipeline/config/runpod.yaml")
+print("  2. Add endpoint_id to 'ultravox' section")
+print("  3. Start the runpod_llm service")
+print("  4. Test via service API")
+print()
+print("Note: vLLM supports TEXT input only")
+print("For AUDIO input, build custom Docker image")
+print()