Separate model loading from CPU warmup

app.py

@@ -327,10 +327,10 @@ def _run_cpu_smoke(model_filter):
         result["phases"]["download"] = {"ok": False, "elapsed_sec": round(time.time() - t0, 2), "error": str(exc)}
         return result
 
-    command = [cli, "-m", path, "-t", str(result["threads"]), "-ngl", "0", "-c", "256", "-n", "1", "-p", "Say hi.", "--no-display-prompt"]
+    command = [cli, "-m", path, "-t", str(result["threads"]), "-ngl", "0", "-c", "256", "-n", "1", "-p", "Say hi.", "--no-display-prompt", "--no-warmup"]
     t0 = time.time()
     try:
-        run = subprocess.run(command, capture_output=True, text=True, timeout=45, env=env)
+        run = subprocess.run(command, capture_output=True, text=True, timeout=180, env=env)
         result["phases"]["load_and_one_token"] = {
             "ok": run.returncode == 0,
             "elapsed_sec": round(time.time() - t0, 2),