Add benchmark test endpoint to diagnose issues

app.py

@@ -626,6 +626,37 @@ def _patched_create_app(blocks, **kwargs):
         except Exception as e:
             return JSONResponse({"error": str(e)}, status_code=500)
 
+    @fa_app.get("/respite/benchmark/test")
+    def _benchmark_test():
+        """Quick test: verify llama.cpp binary works."""
+        import subprocess as _sp
+        try:
+            llama_cli = _ensure_llama_cpp()
+            # Test binary
+            r = _sp.run([llama_cli, "--version"], capture_output=True, text=True, timeout=10)
+            binary_ok = r.returncode == 0
+            binary_info = r.stdout.strip()[:200] if binary_ok else r.stderr.strip()[:200]
+            # Test model download
+            from huggingface_hub import hf_hub_download
+            model_path = hf_hub_download(
+                repo_id="unsloth/gemma-4-E4B-it-GGUF",
+                filename="gemma-4-E4B-it-Q4_K_M.gguf",
+                cache_dir=_BENCHMARK_CACHE,
+            )
+            model_ok = os.path.isfile(model_path)
+            model_size = os.path.getsize(model_path) if model_ok else 0
+            return JSONResponse({
+                "binary_path": llama_cli,
+                "binary_exists": os.path.isfile(llama_cli),
+                "binary_ok": binary_ok,
+                "binary_info": binary_info,
+                "model_path": model_path,
+                "model_ok": model_ok,
+                "model_size_gb": round(model_size / 1024**3, 2),
+            })
+        except Exception as e:
+            return JSONResponse({"error": str(e)}, status_code=500)
+
     _log("Respite API discovery routes injected")
     return fa_app