fix history shit + 4096 output

app.py

@@ -139,7 +139,7 @@ with gr.Blocks(theme='gradio/monochrome') as demo:
             frequency_penalty = gr.Slider(-2, 2, value=0, step=0.1, label="Frequency Penalty")
             presence_penalty = gr.Slider(-2, 2, value=0, step=0.1, label="Presence Penalty")
             repetition_penalty = gr.Slider(0.01, 5, value=1.1, step=0.01, label="Repetition Penalty")
-            max_tokens = gr.Slider(1, 2048, value=256, step=1, label="Max Output (max_tokens)")
+            max_tokens = gr.Slider(1, 4096, value=256, step=1, label="Max Output (max_tokens)")
 
     def user(user_message, history):
         print(f"{get_timestamp()} <|user|> {user_message}")
@@ -163,13 +163,13 @@ with gr.Blocks(theme='gradio/monochrome') as demo:
         stop_generation.clear()
 
     def regenerate_response(history, system_prompt, temperature, top_p, top_k, frequency_penalty, presence_penalty, repetition_penalty, max_tokens):
-        if len(history) > 0:
+        if history is not None and len(history) > 0:
             last_user_message = history[-1][0]
             history[-1][1] = None  
             for new_history in bot(history, system_prompt, temperature, top_p, top_k, frequency_penalty, presence_penalty, repetition_penalty, max_tokens):
                 yield new_history
         else:
-            yield history
+            yield [] if history is None else history
 
     def import_chat_wrapper(custom_format_string):
         imported_history, imported_system_prompt = import_chat(custom_format_string)
@@ -198,4 +198,4 @@ with gr.Blocks(theme='gradio/monochrome') as demo:
     stop_btn.click(stop_generation_func, inputs=[], outputs=[])
 
 if __name__ == "__main__":
-    demo.launch(debug=True) # print the retardation
+    demo.launch(debug=True)