Update app.py

app.py

@@ -56,12 +56,15 @@ def format_prompt_default(message, history,cust_p):
     if history:
         #<start_of_turn>userHow does the brain work?<end_of_turn><start_of_turn>model
         for user_prompt, bot_response in history:
-            prompt += f"{user_prompt}\n"
+            prompt += f"<|startoftext|>{user_prompt}<|endoftext|>"
             print(prompt)
-            prompt += f"{bot_response}\n"
+            prompt += f"<|startoftext|>{bot_response}<|endoftext|>"
             print(prompt)
-    #prompt += f"{message}\n"
+    prompt += "<|startoftext|>"
     prompt+=cust_p.replace("USER_INPUT",message)
+    prompt += "<|endoftext|>"
+    prompt += "<|startoftext|>"
+    
     return prompt
 
 
@@ -72,7 +75,7 @@ def load_models(inp):
     print(models[inp])
     model_state= InferenceClient(models[inp])
     out_box=gr.update(label=models[inp])
-    prompt_out="USER_INPUT\n"
+    prompt_out="<|startoftext|>USER_INPUT<|endoftext|>"
     return out_box,prompt_out, model_state
     
 
@@ -167,9 +170,9 @@ with gr.Blocks() as app:
                     rand = gr.Checkbox(label="Random Seed", value=True)
                     seed=gr.Slider(label="Seed", minimum=1, maximum=1111111111111111,step=1, value=rand_val)
                     tokens = gr.Slider(label="Max new tokens",value=1600,minimum=0,maximum=8000,step=64,interactive=True, visible=True,info="The maximum number of tokens")
-                    temp=gr.Slider(label="Temperature",step=0.01, minimum=0.01, maximum=1.0, value=0.49)
-                    top_p=gr.Slider(label="Top-P",step=0.01, minimum=0.01, maximum=1.0, value=0.49)
-                    rep_p=gr.Slider(label="Repetition Penalty",step=0.01, minimum=0.1, maximum=2.0, value=0.99)
+                    temp=gr.Slider(label="Temperature",step=0.01, minimum=0.01, maximum=1.0, value=0.99)
+                    top_p=gr.Slider(label="Top-P",step=0.01, minimum=0.01, maximum=1.0, value=0.99)
+                    rep_p=gr.Slider(label="Repetition Penalty",step=0.01, minimum=0.1, maximum=2.0, value=1.2)
                     chat_mem=gr.Number(label="Chat Memory", info="Number of previous chats to retain",value=4)
         with gr.Accordion(label="Screenshot",open=False):
             with gr.Row():