Update app.py

app.py

@@ -52,7 +52,7 @@ h3 {
     text-align: center;
 }
 """
-def stream_chat(message: str, history: list, temperature: float, context_window: int, top_p: float, top_k: int, penalty: float):
+def stream_chat(message: str, history: list, temperature: float, max_new_tokens: int, top_p: float, top_k: int, penalty: float):
     print(f'message is - {message}')
     print(f'history is - {history}')
     conversation = []
@@ -69,11 +69,13 @@ def stream_chat(message: str, history: list, temperature: float, context_window:
         model="gemma2",
         messages=conversation,
         stream=True,
-        temperature=temperature,
-        top_k=top_k,
-        top_p=top_p,
-        num_ctx=context_window,
-        repeat_penalty=penalty,
+        options={
+            'num_predict': max_new_tokens,
+            'temperature': temperature,
+            'top_p': top_p,
+            'top_k': top_k,
+            'repeat_penalty': penalty,
+        },
     )
 
     buffer = ""
@@ -108,7 +110,7 @@ with gr.Blocks(css=CSS, theme="soft") as demo:
                 maximum=2048,
                 step=1,
                 value=1024,
-                label="Context window",
+                label="Max New Tokens",
                 render=False,
             ),
             gr.Slider(