Update app.py

app.py

@@ -4,21 +4,25 @@ from huggingface_hub import InferenceClient
 client = InferenceClient("Futuresony/future_ai_12_10_2024.gguf")
 
 def respond(message, history, max_tokens, temperature, top_p):
-    formatted_prompt = message.strip()  # Ensure clean input
+    message = message.strip()  # Ondoa nafasi zisizo za lazima
+    history = history[-5:]  # Chukua historia ya mwisho tu (kuepuka kujirudia)
 
     response = client.text_generation(
-        formatted_prompt, 
-        max_new_tokens=int(max_tokens),  # ✅ Ensure integer
+        message,
+        max_new_tokens=int(max_tokens),
         temperature=float(temperature),
         top_p=float(top_p),
     )
 
-    yield response.strip()  # ✅ Clean output
+    # Safisha majibu, ondoa sehemu zisizo za lazima
+    response = response.replace("Instruction:", "").replace("Response:", "").strip()
+
+    return response
 
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
-        gr.Slider(minimum=1, maximum=250, value=128, step=1, label="Max new tokens"),  # ✅ Ensure integer input
+        gr.Slider(minimum=1, maximum=250, value=128, step=1, label="Max new tokens"),
         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
         gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p"),
     ],