Update app.py

app.py

@@ -32,18 +32,19 @@ def load_model():
         torch_dtype=torch.float32
     )
 
-    # Load LoRA from /lora if it exists
+    # Load LoRA from /lora (FAIL if missing — no silent fallback)
     model = PeftModel.from_pretrained(
-    base_model,
-    f"{MODEL_ID}/lora",
-    is_trainable=False
+        base_model,
+        f"{MODEL_ID}/lora",
+        is_trainable=False
     )
 
-print("LoRA loaded")
-print(model.peft_config)
+    print(model.peft_config)
 
+    model.set_adapter("default")  # ensure adapter is active
     model.to("cpu")
     model.eval()
+
     return tokenizer, model