f

app.py

@@ -28,29 +28,32 @@ model = PeftModel.from_pretrained(
     ignore_mismatched_sizes=True,
 )
 
-def classify_food(text):
+def extract_food_term(text):
     """
-    Classify or extract food-related terms from the input text.
+    Extract or simplify a food term to a single word or best descriptor.
     """
-    prompt = f"Below is some text. Please identify and classify food-related terms.\nText: {text}\nFood classification or extraction:"
+    prompt = f"Extract the best single word or term that describes this food item:\nInput: {text}\nOutput:"
     inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
     with torch.no_grad():
         outputs = model.generate(
             **inputs,
-            max_new_tokens=64,
-            temperature=0.7,  # Adjust temperature for creativity
-            top_p=0.9,       # Adjust top_p for diversity
+            max_new_tokens=16,  # Limit output to a single word
+            temperature=0.0,   # Deterministic output
+            top_p=1.0,         # Focus on the most likely word
+            do_sample=False,   # Disable sampling
         )
     answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    # Extract only the generated output
+    answer = answer.split("Output:")[1].strip()
     return answer
 
 with gr.Blocks() as demo:
-    gr.Markdown("## Qwen + LoRA Adapter: Food Classification/Extraction Demo")
-    input_box = gr.Textbox(lines=3, label="Enter text containing food items")
-    output_box = gr.Textbox(lines=3, label="Model's classification or extraction output")
+    gr.Markdown("## Qwen + LoRA Adapter: Food Term Extraction Demo")
+    input_box = gr.Textbox(lines=1, label="Enter a food item (e.g., 'blaubeertorte')")
+    output_box = gr.Textbox(lines=1, label="Best single-word descriptor")
 
-    classify_btn = gr.Button("Analyze Food Terms")
-    classify_btn.click(fn=classify_food, inputs=input_box, outputs=output_box)
+    extract_btn = gr.Button("Extract Term")
+    extract_btn.click(fn=extract_food_term, inputs=input_box, outputs=output_box)
 
 if __name__ == "__main__":
     demo.launch()