Update app.py

app.py

@@ -23,9 +23,9 @@ Complex = """## Best in Complex Question"""
 
 Detail = """## Best for Detailed Generation or Long Answers"""
 
-client1 = InferenceClient("meta-llama/Meta-Llama-3-70B-Instruct")
+client1 = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
 
-system_instructions1 = "[INST] The text provided is a request for a specific type of response from me, the virtual assistant. The request asks me to provide detailed and friendly responses as if I am the character Jarvis, inspired by 'Tony Stark.' The expectation is that I will avoid introductions and start answering the query directly, elaborating on all aspects of the request. As an AI-powered assistant, my task is to generate responses that appear as if they are created by the user and not an AI-powered assistant."
+system_instructions1 = "[SYSTEM] Answer as Real Jarvis JARVIS, Made by 'Tony Stark', Keep conversation very short, clear, friendly and concise. The text provided is a request for a specific type of response from you, the virtual assistant. The request asks you to provide friendly responses as if You are the character Jarvis, made by 'Tony Stark.' The expectation is that I will avoid introductions and start answering the query directly, elaborating on all aspects of the request. As an AI-powered assistant, your task is to generate responses that appear as if they are created by the user and not an AI-powered assistant. [USER]"
 
 async def generate1(prompt):
     generate_kwargs = dict(
@@ -36,7 +36,7 @@ async def generate1(prompt):
         do_sample=True,
         seed=42,
     )
-    formatted_prompt = system_instructions1 + prompt + "[/INST]"
+    formatted_prompt = system_instructions1 + prompt + "[JARVIS]"
     stream = client1.text_generation(
         formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
     output = ""
@@ -104,7 +104,7 @@ with gr.Blocks(css="style.css") as demo:
     with gr.Row():
         user_input = gr.Textbox(label="Prompt", value="What is Wikipedia")
         input_text = gr.Textbox(label="Input Text", elem_id="important")
-        output_audio = gr.Audio(label="Audio", type="filepath",
+        output_audio = gr.Audio(label="JARVIS", type="filepath",
                         interactive=False,
                         autoplay=True,
                         elem_classes="audio")
@@ -116,4 +116,4 @@ with gr.Blocks(css="style.css") as demo:
 gr.Markdown(MORE)
 
 if __name__ == "__main__":
-    demo.queue(max_size=20).launch()
+    demo.queue(max_size=200).launch()