Update app.py

app.py

@@ -101,15 +101,16 @@ def chat_fn(user_input, history):
 
     context = retrieve_context(user_input)
 
-    # 🔥 If no context → strict fallback
+    # No context
     if not context:
         response = "I don't know."
+
     else:
-        # 🔥 If context is small → return directly (FAST RAG)
+        # Small context → direct RAG
         if len(context.split()) < 50:
             response = context.strip()
+
         else:
-            # 🔥 Use LLM with strict instruction
             prompt = f"""
 You are a livestock expert assistant for goats and cows.
 
@@ -125,16 +126,24 @@ Question:
 Answer in short and clear sentences.
 """
 
-            output = generator(prompt)
+            output = generator(
+                prompt,
+                max_new_tokens=120,   # ✅ remove max_length warning
+                do_sample=True,
+                temperature=0.6
+            )
+
             text = output[0]["generated_text"]
 
-            # Clean output
             if prompt.strip() in text:
                 text = text.split(prompt.strip())[-1].strip()
 
             response = text
 
-    history.append((user_input, response))
+    # ✅ FIXED FORMAT (IMPORTANT)
+    history.append({"role": "user", "content": user_input})
+    history.append({"role": "assistant", "content": response})
+
     return history
 
 # -----------------------------
@@ -143,8 +152,8 @@ Answer in short and clear sentences.
 with gr.Blocks() as demo:
     gr.Markdown("## 🐐 Livestock Chatbot (RAG + Qwen)")
 
-    chatbot = gr.Chatbot()
-    msg = gr.Textbox(label="Ask about goats or cows")
+    chatbot = gr.Chatbot(type="messages")  # ✅ REQUIRED
+    msg = gr.Textbox()
     btn = gr.Button("Send")
 
     btn.click(chat_fn, [msg, chatbot], chatbot)