fix: remove top-k sampling

app.py

@@ -12,8 +12,7 @@ def respond(
     history: list[tuple[str, str]],
     max_tokens: int,
     temperature: float,
-    top_p: float,
-    top_k: int
+    top_p: float
 ):
     messages = []
 
@@ -32,8 +31,7 @@ def respond(
         max_tokens=max_tokens,
         stream=True,
         temperature=temperature,
-        top_p=top_p,
-        top_k=top_k
+        top_p=top_p
     ):
         token = message.choices[0].delta.content
 
@@ -56,13 +54,6 @@ demo = gr.ChatInterface(
             step=0.05,
             label="Top-p",
         ),
-        gr.Slider(
-            minimum=10,
-            maximum=100,
-            value=30,
-            step=10,
-            label="Top-k"
-        ),
     ],
 )