Update app.py

app.py

@@ -33,21 +33,28 @@ def _gpu_duration_gallery(images: List[Image.Image], *_, **__) -> int:
     return min(600, 45 * n + 60)
 
 def _load(model_id: str):
-    # Use env HF_TOKEN for gated access per hub docs
-    processor = AutoImageProcessor.from_pretrained(model_id, token=HF_TOKEN if HF_TOKEN else None)
+    # token works on current Transformers
+    processor = AutoImageProcessor.from_pretrained(
+        model_id,
+        use_fast=True,
+        token=HF_TOKEN if HF_TOKEN else None,
+    )
     model = AutoModel.from_pretrained(
         model_id,
         torch_dtype=torch.float16,
         low_cpu_mem_usage=True,
         token=HF_TOKEN if HF_TOKEN else None,
-    )
-    model.to("cuda").eval()
+    ).to("cuda").eval()
     return processor, model
 
 def _extract_core(image: Image.Image, model_id: str, pooling: str, want_overlay: bool):
     t0 = time.time()
     processor, model = _load(model_id)
-    inputs = processor(images=image, return_tensors="pt").to("cuda")
+
+    # safer move to cuda for BatchFeature
+    inputs = processor(images=image, return_tensors="pt")
+    inputs = {k: v.to("cuda") for k, v in inputs.items()}
+
     with torch.cuda.amp.autocast(dtype=torch.float16), torch.inference_mode():
         out = model(**inputs)