Update app.py

app.py

@@ -264,15 +264,15 @@ def generate(
     )
 
     formatted_prompt = format_prompt(f"{system_prompt}, {prompt}", history)
-    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 output
+    output = client.text_generation(formatted_prompt, **generate_kwargs, stream=False, return_full_text=True)
     
     return output
 
+def chat_interface(prompt, history, agent_name, sys_prompt, temperature, max_new_tokens, top_p, repetition_penalty):
+    generated_text = generate(prompt, history, agent_name, sys_prompt, temperature, max_new_tokens, top_p, repetition_penalty)
+    chatbot_response = [(prompt, generated_text)]
+    return chatbot_response, ""  # Return the chatbot response and an empty string for the message textbox
+
 additional_inputs = [
     gr.Dropdown(
         label="Agents",
@@ -343,13 +343,6 @@ def create_interface():
         )
     return iface
 
-def chat_interface(prompt, history, agent_name, sys_prompt, temperature, max_new_tokens, top_p, repetition_penalty):
-    generated_text = generate(prompt, history, agent_name, sys_prompt, temperature, max_new_tokens, top_p, repetition_penalty)
-    chatbot_response = []
-    for response in generated_text:
-        chatbot_response.append([prompt, response])
-    return chatbot_response, ""  # Return the chatbot response and an empty string for the message textbox
-
 iface = gr.Blocks()
 
 with iface: