Fix: install llama-cpp-python at runtime instead of build time

app.py

@@ -420,10 +420,17 @@ def _run_benchmark():
     if _BENCHMARK_RESULTS:
         return _BENCHMARK_RESULTS
 
+    # Install llama-cpp-python at runtime (not in requirements to avoid build issues)
     try:
         from llama_cpp import Llama
     except ImportError:
-        return [{"error": "llama-cpp-python not installed"}]
+        _log("Benchmark: installing llama-cpp-python...")
+        import subprocess
+        subprocess.check_call(
+            [sys.executable, "-m", "pip", "install", "llama-cpp-python>=0.4.0", "--only-binary=:all:"],
+            stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL,
+        )
+        from llama_cpp import Llama
 
     os.makedirs(_BENCHMARK_CACHE, exist_ok=True)
     results = []