changes

app.py

@@ -140,6 +140,67 @@ def generate_answer(image, question, temperature=0.7, top_p=0.95, max_tokens=256
     trimmed = [o[len(i):] for i, o in zip(inputs["input_ids"], out)]
     return processor.batch_decode(trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
 
+# ---- PetCare answer + product suggestions (ONE output) ----
+@spaces.GPU(duration=120)
+def pet_answer_with_recs(image, question, temperature=0.7, top_p=0.95, max_tokens=256, budget_twd=None):
+    """
+    1) Get the normal PetBull answer (image + text).
+    2) Run vector search on the user's question.
+    3) Ask the LLM (text-only) to decide if any candidates are relevant for the user's issue.
+       If yes, append a 'Suggested products' section (up to 3 items from the list).
+       If not, append 'No relevant products.'.
+    """
+    # Step 1: normal PetBull answer
+    base = generate_answer(image, question, temperature, top_p, max_tokens)
+
+    # Step 2: retrieve product candidates (humans/skincare; model will decide relevance)
+    cands = product_search(question, k=8)
+    cand_block = format_candidates_for_llm(cands, budget_twd=budget_twd)
+
+    # Step 3: build a small, text-only prompt for suggestions
+    # IMPORTANT: we use the same Qwen2.5-VL model in text mode
+    messages = [{
+        "role": "user",
+        "content": [
+            {"type": "text", "text":
+             "You are DermalCare's assistant.\n"
+             "Task: Given a user message and a list of candidate skincare products, "
+             "return a section titled 'Suggested products' with up to 3 items ONLY IF they are relevant to the user's issue. "
+             "Each item must be strictly chosen from the candidate list; include one-line 'why it helps' and a short 'how to use'. "
+             "If none are relevant (e.g., purely veterinary emergencies, non-skin issues), output exactly: 'No relevant products.'"}
+        ]
+    },{
+        "role": "user",
+        "content": [
+            {"type": "text", "text": f"User message:\n{question}"},
+            {"type": "text", "text": f"Candidates (do not invent beyond this list):\n{cand_block}"}
+        ]
+    }]
+
+    # Prepare inputs on GPU (text-only)
+    text_prompt = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
+    inputs = processor(
+        text=[text_prompt], images=None, videos=None, padding=True, return_tensors="pt",
+    )
+    inputs = {k: (v.to("cuda") if hasattr(v, "to") else v) for k, v in inputs.items()}
+
+    with torch.no_grad():
+        out = model.generate(
+            **inputs,
+            max_new_tokens=200,
+            temperature=0.2,      # keep precise/grounded
+            top_p=0.95,
+        )
+    trimmed = [o[len(i):] for i, o in zip(inputs["input_ids"], out)]
+    suggestions = processor.batch_decode(trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+
+    # Final combined message
+    safety = (
+        "Safety notes: For broken/infected skin, pregnancy/lactation, infants, "
+        "or if symptoms worsen—seek a qualified dermatologist. Patch-test first."
+    )
+    return f"{base}\n\n{('Suggested products:\n' + suggestions) if suggestions.strip() != '' else ''}\n\n{safety}"
+
 # ---- UI ----
 with gr.Blocks(title="DermalCare - Pet & Skincare Assistant") as demo:
     gr.Markdown("# DermalCare - Your AI Assistant for Pet Care and Skincare")
@@ -158,14 +219,10 @@ with gr.Blocks(title="DermalCare - Pet & Skincare Assistant") as demo:
                 with gr.Column():
                     answer = gr.Textbox(lines=12, label="Assistant", interactive=False)
 
-            ask.click(generate_answer, inputs=[img_in, txt_in, temp, topp, max_tok], outputs=answer)
-        
-        with gr.Tab("Product Recommender"):
-            need = gr.Textbox(label="Describe your need (中文或English)", lines=3,
-                              value="敏感肌 想找溫和溫和拔乾的保濕潔面，預算 < TWD 400")
-            budget = gr.Number(label="Budget (TWD, optional)", value=400, precision=0)
-            btn = gr.Button("Recommend", variant="primary")
-            out = gr.Textbox(label="Recommendations", lines=14)
-            btn.click(fn=recommend_products, inputs=[need, budget], outputs=out, queue=True)
+            ask.click(
+                pet_answer_with_recs,
+                inputs=[img_in, txt_in, temp, topp, max_tok],   # (budget optional: add at the end if you want)
+                outputs=answer
+            )
 
 demo.queue().launch(show_api=False, share=True)