remove autocast

app.py

@@ -101,12 +101,7 @@ def detect(
     inputs = bundle.processor(images=image, text=texts, return_tensors="pt").to(DEVICE)
 
     with torch.inference_mode():
-        if DEVICE == "cuda":
-            # Use autocast to speed up mixed-precision-friendly ops
-            with torch.amp.autocast():
-                outputs = bundle.model(**inputs)
-        else:
-            outputs = bundle.model(**inputs)
+        outputs = bundle.model(**inputs)
 
     results = bundle.processor.post_process_grounded_object_detection(
         outputs, threshold=threshold, target_sizes=[image.size[::-1]]