Update app.py

app.py

@@ -54,15 +54,15 @@ def format_prompt(message, history):
 result = []
 
 
-def chat_inf(system_prompt,prompt,history,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem):
+def chat_inf(system_prompt,prompt,history,memory,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem):
     #token max=8192
     hist_len=0
     client=clients[int(client_choice)-1]
     if not history:
         history = []
         hist_len=0
-    if history:
-        for ea in history[0-chat_mem:]:
+    if memory:
+        for ea in memory[0-chat_mem:]:
             hist_len+=len(str(ea))
         print(hist_len)
     in_len=len(system_prompt+prompt)+hist_len
@@ -71,7 +71,8 @@ def chat_inf(system_prompt,prompt,history,client_choice,seed,temp,tokens,top_p,r
     print("\n######### HIST "+str(in_len))
     print("\n######### TOKENS "+str(tokens))
     if (in_len+tokens) > 8000:
-        yield [(prompt,"Wait, that's too many tokens, please reduce the Chat Memory value")]
+        history.append((prompt,"Wait, that's too many tokens, please reduce the Chat Memory value"))
+        yield history,memory
         #hist=compress_history(history,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem)
         #yield [(prompt,"History has been compressed, processing request...")]
         #history.append((prompt,hist))
@@ -85,16 +86,16 @@ def chat_inf(system_prompt,prompt,history,client_choice,seed,temp,tokens,top_p,r
             seed=seed,
         )
         #formatted_prompt=prompt   
-        formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", history[0-chat_mem:])
+        formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", memory[0-chat_mem:])
         print("\n######### PROMPT "+str(len(formatted_prompt)))
         stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
         output = ""
         for response in stream:
             output += response.token.text
-            yield [(prompt,output)]
+            yield [(prompt,output)],memory
         history.append((prompt,output))
-        memory=history
-        yield history
+        memory.append((prompt,output))
+        yield history,memory
 
 def get_screenshot(chat: list,height=5000,width=600,chatblock=[],theme="light",wait=3000,header=True):
     print(chatblock)
@@ -162,9 +163,9 @@ with gr.Blocks() as app:
 
 
     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,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem],chat_b)
+    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,client_choice,seed,temp,tokens,top_p,rep_p,chat_mem],[chat_b])
+    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])
     
     stop_btn.click(None,None,None,cancels=[go,im_go,chat_sub])
     clear_btn.click(clear_fn,None,[inp,sys_inp,chat_b])