fix: update LAST_METRICS type annotation and remove unused demo.load call

app.py

@@ -46,7 +46,7 @@ logging.basicConfig(level=LOG_LEVEL, format="%(asctime)s %(levelname)s %(message
 logger = logging.getLogger("zamai-app")
 
 # Metrics storage for last real generation
-LAST_METRICS = {
+LAST_METRICS: dict[str, float | int | str | None] = {
     "latency_sec": None,
     "input_tokens": None,
     "output_tokens": None,
@@ -164,7 +164,6 @@ def predict(instruction: str,
         if input_text and input_text.strip():
             return base + "\n" + input_text.strip()
         return base
-            demo.load(_post_load, inputs=None, outputs=status_box)
 
     prompt = build_prompt()
     active_mode = (mode or "").strip().lower() or ECHO_MODE