Update app.py

app.py

@@ -50,8 +50,8 @@ model = OlmoeForCausalLM.from_pretrained(
 def stream_chat(
     message: str, 
     history: list, 
-    temperature: float = 0.3, 
-    max_new_tokens: int = 1024, 
+    temperature: float = 0.5, 
+    max_new_tokens: int = 256, 
     top_p: float = 1.0, 
     top_k: int = 20, 
     penalty: float = 1.2,
@@ -110,15 +110,15 @@ with gr.Blocks(css=CSS, theme="Nymbo/Nymbo_Theme") as demo:
                 minimum=0,
                 maximum=1,
                 step=0.1,
-                value=0.3,
+                value=0.5,
                 label="Temperature",
                 render=False,
             ),
             gr.Slider(
                 minimum=128,
-                maximum=8192,
+                maximum=4096,
                 step=1,
-                value=1024,
+                value=256,
                 label="Max new tokens",
                 render=False,
             ),