Add explicit exit code and model verification logging

app.py

@@ -507,16 +507,23 @@ def _run_benchmark():
                 "--no-display-prompt",
                 "--log-disable",
             ]
+            _log(f"Benchmark: cmd: {' '.join(cmd[:6])}...")
+            _log(f"Benchmark: model exists: {os.path.isfile(model_path)}, size: {os.path.getsize(model_path) if os.path.isfile(model_path) else 0}")
             t0 = time.time()
-            proc = subprocess.run(
-                cmd, capture_output=True, text=True, timeout=300,
-            )
+            try:
+                proc = subprocess.run(
+                    cmd, capture_output=True, text=True, timeout=300,
+                )
+            except subprocess.TimeoutExpired:
+                result["error"] = "Timeout after 300s"
+                results.append(result)
+                continue
             wall_time = time.time() - t0
             output_text = proc.stdout.strip()
 
-            # Log raw output for debugging
-            _log(f"Benchmark: {model_info['name']} stdout ({len(output_text)} chars): {output_text[:200]}")
-            _log(f"Benchmark: {model_info['name']} stderr ({len(proc.stderr)} chars): {proc.stderr[-500:]}")
+            _log(f"Benchmark: {model_info['name']} exit code: {proc.returncode}")
+            _log(f"Benchmark: {model_info['name']} stdout ({len(output_text)} chars): {output_text[:300]}")
+            _log(f"Benchmark: {model_info['name']} stderr ({len(proc.stderr)} chars): {proc.stderr[-600:]}")
 
             # Parse llama.cpp stderr for timing info
             # Various formats depending on version: