Add vLLM serve script, systemd service, and validation test for Qwen3.6-35B-A3B-FP8 on dankserver

scripts/validate-vllm.sh

@@ -0,0 +1,93 @@
+#!/usr/bin/env bash
+# End-to-end validation test for vLLM Qwen3.6-35B-A3B-FP8 on dankserver
+# Sends a chat completion request and measures prompt processing speed
+set -euo pipefail
+
+PORT="${PORT:-8000}"
+HOST="${HOST:-127.0.0.1}"
+URL="http://${HOST}:${PORT}"
+MIN_SPEED="${MIN_SPEED:-5.0}"
+MAX_TOKENS="${MAX_TOKENS:-256}"
+
+echo "=== vLLM Validation Test ==="
+echo "Target: ${URL}"
+echo "Min prompt speed: ${MIN_SPEED} tok/s"
+echo ""
+
+# Wait for server to be ready
+echo "Waiting for server..."
+for i in $(seq 1 300); do
+    if curl -sf "${URL}/health" > /dev/null 2>&1; then
+        echo "Server ready after $((i * 2))s"
+        break
+    fi
+    if [ "$i" -eq 300 ]; then
+        echo "FAIL: Server did not become ready in 600s"
+        exit 1
+    fi
+    sleep 2
+done
+
+# Send chat completion and measure prompt processing speed
+echo ""
+echo "--- Sending validation prompt ---"
+RESPONSE=$(curl -s -X POST "${URL}/v1/chat/completions" \
+    -H "Content-Type: application/json" \
+    -d "{
+        \"model\": \"frank-default-large\",
+        \"messages\": [
+            {\"role\": \"user\", \"content\": \"Explain the difference between supervised learning and unsupervised learning in two sentences.\"}
+        ],
+        \"max_tokens\": ${MAX_TOKENS}
+    }")
+
+echo "Response received."
+echo ""
+
+# Parse timing from response
+echo "$RESPONSE" | python3 -c "
+import sys, json
+
+data = json.load(sys.stdin)
+
+# Check for errors
+if 'error' in data:
+    print(f'ERROR: {data[\"error\"]}')
+    sys.exit(1)
+
+# Extract usage info
+usage = data.get('usage', {})
+prompt_tokens = usage.get('prompt_tokens', 0)
+completion_tokens = usage.get('completion_tokens', 0)
+
+print(f'Prompt tokens:   {prompt_tokens}')
+print(f'Completion tokens: {completion_tokens}')
+
+# Extract timing details
+prompt_details = usage.get('prompt_details', {})
+if prompt_details:
+    prompt_time = prompt_details.get('total_time', 0)
+    speed = prompt_tokens / prompt_time if prompt_time > 0 else 0
+    print(f'Prompt time:   {prompt_time:.4f}s')
+    print(f'Prompt speed:  {speed:.1f} tok/s')
+    
+    if speed < float('${MIN_SPEED}'):
+        print(f'FAIL: Speed {speed:.1f} tok/s < {\"${MIN_SPEED}\"} tok/s')
+        sys.exit(1)
+    else:
+        print(f'PASS: Speed {speed:.1f} tok/s >= {\"${MIN_SPEED}\"} tok/s')
+else:
+    # Fallback: use total_time from usage
+    total_time = usage.get('total_time', 0)
+    if total_time > 0 and prompt_tokens > 0:
+        speed = prompt_tokens / total_time
+        print(f'Total time:    {total_time:.4f}s')
+        print(f'Est. prompt speed: {speed:.1f} tok/s')
+        print('PASS (response received)')
+    else:
+        print('No timing details available')
+        print('PASS (response received)')
+"
+
+echo ""
+echo "=== Validation complete ==="