Update app.py

app.py

@@ -91,7 +91,7 @@ def extract_json(text: str) -> Dict[str, Any]:
     return {"raw_output": text}
 
 
-PROMPT = "Analyze this pantry image in detail, list every ingredient"
+PROMPT = "Analyze this pantry image in detail, list every ingredient as JSON"
 
 
 
@@ -108,7 +108,7 @@ def analyze_pantry(image: Image.Image) -> Tuple[Image.Image, Dict[str, Any]]:
         {
             "role": "system",
             "content": [
-                {"type": "text", "text": "return JSON only."}
+                {"type": "text", "text": "Analyze this pantry image in detail, list every ingredient"}
             ],
         },
         {
@@ -130,6 +130,7 @@ def analyze_pantry(image: Image.Image) -> Tuple[Image.Image, Dict[str, Any]]:
     )
 
     inputs = inputs.to(model.device)
+    print("inputs:", inputs)
 
     with torch.inference_mode():
         output_ids = model.generate(
@@ -145,6 +146,8 @@ def analyze_pantry(image: Image.Image) -> Tuple[Image.Image, Dict[str, Any]]:
         clean_up_tokenization_spaces=False,
     )[0].strip()
 
+    print("generated_text:", generated_text)
+
     parsed = extract_json(generated_text)
     if isinstance(parsed, dict) and "raw_output" not in parsed:
         parsed["_raw_output"] = generated_text
@@ -154,9 +157,6 @@ def analyze_pantry(image: Image.Image) -> Tuple[Image.Image, Dict[str, Any]]:
 
 with gr.Blocks() as demo:
     gr.Markdown("# Pantry Scanner")
-    gr.Markdown(
-        "Single-image Qwen3-VL OCR/brand reader. No tiling, no crop pipeline, no manual sharpening."
-    )
 
     with gr.Row():
         image_input = gr.Image(type="pil", label="Pantry image")