Added top_p

app.py

@@ -7,11 +7,12 @@ import requests
 title = "A conversation with Gandalf (GPTJ-6B) 🧙"
 description = ""
 article = """
-<p> To reset you need to reload the page </p>
+<p> To reset you <b>need to reload the page.</b> </p>
 <h2> Parameters: </h2>
 <ul>
-    <li>temperature: (sampling temperature) higher values means the model will take more risks.</li>
-    <li>max_new_tokens: Max number of tokens in generation.</li>
+    <li><i>top_p</i>:  control how deterministic the model is in generating a response.</li>
+    <li><i>temperature</i>: (sampling temperature) higher values means the model will take more risks.</li>
+    <li><i>max_new_tokens</i>: Max number of tokens in generation.</li>
 </ul>
 <img src='http://www.simoninithomas.com/test/gandalf.jpg', alt="Gandalf"/>"""
 theme="huggingface"
@@ -45,7 +46,7 @@ def clean_chat_output(txt, prompt):
   return output
 
 
-def chat(top_p, max_new_tokens, message):
+def chat(top_p, temperature, max_new_tokens, message):
     history = gr.get_state() or []
     history.append((message, ""))
     gr.set_state(history)
@@ -56,6 +57,7 @@ def chat(top_p, max_new_tokens, message):
     json_ = {"inputs": prompt,
          "parameters":
          {
+         "top_p": top_p,
         "temperature": temperature,
         "max_new_tokens": max_new_tokens,
         "return_full_text": False
@@ -77,6 +79,7 @@ def chat(top_p, max_new_tokens, message):
 iface = gr.Interface(
         chat, 
         [ 
+            gr.inputs.Slider(minimum=0.5, maximum=1, step=0.05, default=0.9, label="top_p"),
             gr.inputs.Slider(minimum=0.5, maximum=1.5, step=0.1, default=1.1, label="temperature"),
             gr.inputs.Slider(minimum=20, maximum=250, step=10, default=50, label="max_new_tokens"),
             "text",