feat: add top-k sampling

app.py

@@ -10,9 +10,10 @@ client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 def respond(
     message,
     history: list[tuple[str, str]],
-    max_tokens,
-    temperature,
-    top_p,
+    max_tokens: int,
+    temperature: float,
+    top_p: float,
+    top_k: int
 ):
     messages = []
 
@@ -32,6 +33,7 @@ def respond(
         stream=True,
         temperature=temperature,
         top_p=top_p,
+        top_k=top_k
     ):
         token = message.choices[0].delta.content
 
@@ -52,7 +54,14 @@ demo = gr.ChatInterface(
             maximum=1.0,
             value=0.95,
             step=0.05,
-            label="Top-p (nucleus sampling)",
+            label="Top-p",
+        ),
+        gr.Slider(
+            minimum=10,
+            maximum=100,
+            value=30,
+            step=10,
+            label="Top-k"
         ),
     ],
 )