fix: run YOLO on main container, not GPU worker (file access + log visibility)

app.py

@@ -231,47 +231,17 @@ def analyze_frame(image: Image.Image, hive_name: str):
 
 
 def _run_qwen(image: Image.Image, prefix: str = "") -> str:
-    """Run Qwen-VL on the image with an optional context prefix. Returns raw text."""
-    model, processor = get_model()
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-    dtype = torch.float16 if device == "cuda" else torch.float32
-    model = model.to(device=device, dtype=dtype)
-
-    messages = [
-        {
-            "role": "user",
-            "content": [
-                {"type": "image", "image": image},
-                {"type": "text", "text": prefix + INSPECTION_PROMPT},
-            ],
-        }
-    ]
-    inputs = processor.apply_chat_template(
-        messages,
-        tokenize=True,
-        add_generation_prompt=True,
-        return_dict=True,
-        return_tensors="pt",
-    ).to(device)
-
-    with torch.no_grad():
-        generated = model.generate(
-            **inputs, max_new_tokens=300, do_sample=False
-        )
-    prompt_len = inputs["input_ids"].shape[1]
-    return processor.batch_decode(
-        generated[:, prompt_len:], skip_special_tokens=True
-    )[0].strip()
+    """Convenience wrapper that calls the GPU-only Qwen function."""
+    return _qwen_only(image, prefix + INSPECTION_PROMPT)
 
 
-@gpu
 def run_comparison(image: Image.Image):
     """A/B: Apiarist (YOLO + Qwen) vs raw Qwen alone, side-by-side."""
     if image is None:
         msg = "Upload a frame photo to compare."
         return None, msg, None, msg
 
-    # ----- LEFT: Apiarist (specialist + generalist) -----
+    # YOLO on main container (CPU, fast)
     detections, annotated = detector.detect(image)
     counts = detector.summarize_counts(detections)
     yolo_active = detector.is_available()