Update app.py

app.py

@@ -17,9 +17,9 @@ InferenceClient(models[2]),
 InferenceClient(models[3]),
 ]
 
-VERBOSE=True
+VERBOSE=False
 
-def format_prompt(message, history):
+def format_prompt(message, history, cust_p):
     prompt = ""
     if history:
         #<start_of_turn>userHow does the brain work?<end_of_turn><start_of_turn>model
@@ -28,10 +28,15 @@ def format_prompt(message, history):
             #print(prompt)
             prompt += f"{bot_response}\n"
             #print(prompt)
-    prompt += f"<start_of_turn>user\n{message}<end_of_turn>\n<start_of_turn>model"
+    #prompt += f"<start_of_turn>user\n{message}<end_of_turn>\n<start_of_turn>model\n"
+    prompt+=cust_p.format(USER_INPUT=message)
     return prompt
 
-def chat_inf(system_prompt,prompt,history,memory,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem):
+def custom_prompt(prompt):
+    return prompt
+
+
+def chat_inf(system_prompt,prompt,history,memory,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem,cust_p):
     #token max=8192
     hist_len=0
     client=clients[int(client_choice)-1]
@@ -58,7 +63,7 @@ def chat_inf(system_prompt,prompt,history,memory,client_choice,seed,temp,tokens,
             do_sample=True,
             seed=seed,
         )
-        formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", memory[0-chat_mem:])
+        formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", memory[0-chat_mem:],cust_p)
         stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
         output = ""
         for response in stream:
@@ -70,7 +75,7 @@ def chat_inf(system_prompt,prompt,history,memory,client_choice,seed,temp,tokens,
     if VERBOSE==True:
         print("\n######### HIST "+str(in_len))
         print("\n######### TOKENS "+str(tokens))        
-        print("\n######### PROMPT "+str(len(formatted_prompt)))
+        #print("\n######### PROMPT "+str(len(formatted_prompt)))
 
 def get_screenshot(chat: list,height=5000,width=600,chatblock=[],theme="light",wait=3000,header=True):
     print(chatblock)
@@ -109,6 +114,7 @@ with gr.Blocks() as app:
                             stop_btn=gr.Button("Stop")
                             clear_btn=gr.Button("Clear")                
                 client_choice=gr.Dropdown(label="Models",type='index',choices=[c for c in models],value=models[0],interactive=True)
+                custom_prompt=gr.Textbox(lines=5,value="<start_of_turn>user\nUSER_INPUT<end_of_turn>\n<start_of_turn>model\n")
             with gr.Column(scale=1):
                 with gr.Group():
                     rand = gr.Checkbox(label="Random Seed", value=True)
@@ -132,8 +138,10 @@ with gr.Blocks() as app:
                     chatblock=gr.Dropdown(label="Chatblocks",info="Choose specific blocks of chat",choices=[c for c in range(1,40)],multiselect=True)
 
     im_go=im_btn.click(get_screenshot,[chat_b,im_height,im_width,chatblock,theme,wait_time],img)
-    chat_sub=inp.submit(check_rand,[rand,seed],seed).then(chat_inf,[sys_inp,inp,chat_b,memory,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem],[chat_b,memory])
-    go=btn.click(check_rand,[rand,seed],seed).then(chat_inf,[sys_inp,inp,chat_b,memory,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem],[chat_b,memory])
+    
+    chat_sub=inp.submit(check_rand,[rand,seed],seed).then(chat_inf,[sys_inp,inp,chat_b,memory,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem,custom_prompt],[chat_b,memory])
+    go=btn.click(check_rand,[rand,seed],seed).then(chat_inf,[sys_inp,inp,chat_b,memory,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem,custom_prompt],[chat_b,memory])
+    
     stop_btn.click(None,None,None,cancels=[go,im_go,chat_sub])
     clear_btn.click(clear_fn,None,[inp,sys_inp,chat_b,memory])
 app.queue(default_concurrency_limit=10).launch()