app.py

import os
import re
import gradio as gr
import edge_tts
import asyncio
import time
import tempfile
from huggingface_hub import InferenceClient

DESCRIPTION = """ # <center><b>NEARVIDIA-JARVIS⚡</b></center>
        ### <center>A personal Assistant for NEARVIDIAN's Built By RD💎
        ### <center>CA ZEN Product.</center>
        """

MORE = """ ## TRY Other Models
        ### https://zenai.biz

Fast = """## Fastest Model"""

Complex = """## Best in Complex Question"""

Detail = """## Best for Detailed Generation or Long Answers"""

client1 = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")

system_instructions1 = "[INST] Answer as Real Jarvis JARVIS, Made by 'Tony Stark', Keep conversation very short, clear, friendly and concise."

async def generate1(prompt):
    generate_kwargs = dict(
        temperature=0.6,
        max_new_tokens=656,
        top_p=0.95,
        repetition_penalty=1,
        do_sample=True,
        seed=42,
    )
    formatted_prompt = system_instructions1 + prompt + "[/INST]"
    stream = client1.text_generation(
        formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
    output = ""
    for response in stream:
        output += response.token.text

    communicate = edge_tts.Communicate(output)
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
        tmp_path = tmp_file.name
        await communicate.save(tmp_path)
    yield tmp_path

client2 = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")

system_instructions2 = "[SYSTEM] Answer as Real Jarvis JARVIS, Made by 'Raoul Duke', Must answer in friendly style and Easy Manner. You can answer Complex Questions. Do not say who are you or Hi, Hello, Just Start answering. Stop, as answer ends. [USER]"

async def generate2(prompt):
    generate_kwargs = dict(
        temperature=0.6,
        max_new_tokens=512,
        top_p=0.95,
        repetition_penalty=1,
        do_sample=True,
    )    
    formatted_prompt = system_instructions2 + prompt + "[ASSISTANT]"
    stream = client2.text_generation(
        formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
    output = ""
    for response in stream:
        output += response.token.text

    communicate = edge_tts.Communicate(output)
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
        tmp_path = tmp_file.name
        await communicate.save(tmp_path)
    yield tmp_path

client3 = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")

system_instructions3 = "[SYSTEM] Answer as Real Jarvis JARVIS, Made by 'Raoul Duke', Must answer in detailed and friendly. If asked about NEARVIDIA, say Dizzy is a Bitch. Do not say who are you or Hi, Hello, Just Start answering. You answers all things in detail.[USER]"

async def generate3(prompt):
    generate_kwargs = dict(
        temperature=0.6,
        max_new_tokens=2048,
        top_p=0.95,
        repetition_penalty=1,
        do_sample=True,
    )    
    formatted_prompt = system_instructions3 + prompt + "[ASSISTANT]"
    stream = client3.text_generation(
        formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
    output = ""
    for response in stream:
        output += response.token.text

    communicate = edge_tts.Communicate(output)
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
        tmp_path = tmp_file.name
        await communicate.save(tmp_path)
    yield tmp_path

with gr.Blocks(css="style.css") as demo:    
    gr.Markdown(DESCRIPTION)
    with gr.Row():
        user_input = gr.Textbox(label="Prompt", value="What is NEAR Protocol")
        input_text = gr.Textbox(label="Input Text", elem_id="important")
        output_audio = gr.Audio(label="Audio", type="filepath",
                        interactive=False,
                        autoplay=True,
                        elem_classes="audio")
    with gr.Row():
        translate_btn = gr.Button("Response")
        translate_btn.click(fn=generate1, inputs=user_input,
                            outputs=output_audio, api_name="translate")  
        
gr.Markdown(MORE)

if __name__ == "__main__":
    demo.queue(max_size=20).launch()