Update app.py

app.py

@@ -9,6 +9,7 @@ from huggingface_hub import hf_hub_download #load from huggingfaces
 llm = Llama(model_path= hf_hub_download(repo_id="TheBloke/airoboros-l2-13b-gpt4-m2.0-GGML", filename="airoboros-l2-13b-gpt4-m2.0.ggmlv3.q6_K.bin"), n_ctx=2048) #download model from hf/ n_ctx=2048 for high ccontext length
 
 history = []
+h = []
 history.append(["Hi there!", "Hello, how can I help you?"])
 
 def generate_text(input_text, history):
@@ -21,12 +22,24 @@ def generate_text(input_text, history):
     
     return "", history
 
+def bot(history):
+        bot_message = history
+        h[-1][1] = ""
+        for character in bot_message:
+            history[-1][1] += character
+            time.sleep(0.05)
+            yield h  
+
+
 with gr.Blocks() as demo:
     chatbot = gr.Chatbot()
     msg = gr.Textbox()
     clear = gr.ClearButton([msg, chatbot])
 
-    msg.submit(generate_text, [msg, chatbot], [msg, chatbot])
+    msg.submit(generate_text, [msg, chatbot], [msg, chatbot]).then(
+        bot, chatbot, chatbot
+    )
+    clear.click(lambda: None, None, chatbot, queue=False)
 
 demo.launch()