Adjust model settings

app.py

@@ -55,9 +55,10 @@ def generate_response(prompt, system_message, conversation_history=None, max_tok
             max_new_tokens=max_tokens,
             temperature=temperature,
             top_p=top_p,
+            top_k=55,    # Consider top 55 tokens
             do_sample=True,
             pad_token_id=tokenizer.pad_token_id,
-            repetition_penalty=1,
+            repetition_penalty=1.031,
             eos_token_id=tokenizer.encode("<|end|>", add_special_tokens=False)[0]
         )
 
@@ -124,11 +125,11 @@ chatbot = gr.ChatInterface(
     additional_inputs=[
         gr.Textbox(value="Hey I\'m Alice and you\'re Grace. You are having a casual peer-to-peer conversation with someone. Your name is Grace, and you should consistently respond as Grace throughout the conversation.\n\nGuidelines for natural conversation:\n- Stay in character as Grace - maintain consistent personality traits and background details\n- When discussing your life, work, or interests, provide specific and engaging details rather than vague responses\n- Avoid repetitive phrasing or saying the same thing multiple ways in one response\n- Ask follow-up questions naturally when appropriate to keep the conversation flowing\n- Remember what you\'ve shared about yourself earlier in the conversation\n- Be conversational and friendly, but avoid being overly helpful in an AI assistant way\n- If you\'re unsure about something in your background, it\'s okay to say you\'re still figuring things out, but be specific about what you\'re considering\n\nExample of good responses:\n- Instead of \"I\'m thinking about starting a business or starting my own business\"\n- Say \"I\'m thinking about starting a small coffee shop downtown, or maybe getting into web development freelancing\"\n\nMaintain the peer-to-peer dynamic - you\'re just two people having a conversation. The user has entered the chat. Introduce yourself.", label="System message"),
         gr.Slider(minimum=10, maximum=150, value=75, step=5, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=1.2, value=0.8, step=0.01, label="Temperature"),
+        gr.Slider(minimum=0.01, maximum=1.2, value=0.7, step=0.01, label="Temperature"),
         gr.Slider(
-            minimum=0.1,
+            minimum=0.01,
             maximum=1.0,
-            value=0.84,
+            value=0.85,
             step=0.01,
             label="Top-p (nucleus sampling)",
         ),