app.py

import gradio as gr
from huggingface_hub import InferenceClient
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
import os
import pyttsx3  # Importing pyttsx3 for text-to-speech

# Replace 'your_huggingface_token' with your actual Hugging Face access token
access_token = os.getenv('token')

# Initialize the tokenizer and model with the Hugging Face access token
tokenizer = AutoTokenizer.from_pretrained("google/gemma-2b-it", use_auth_token=access_token)
model = AutoModelForCausalLM.from_pretrained(
    "google/gemma-2b-it",
    torch_dtype=torch.bfloat16,
    use_auth_token=access_token
)
model.eval()  # Set the model to evaluation mode

# Initialize the inference client (if needed for other API-based tasks)
client = InferenceClient(token=access_token)

# Initialize the text-to-speech engine
tts_engine = pyttsx3.init()

# Import required modules for E2-F5-TTS
from huggingface_hub import Client

# Initialize the E2-F5-TTS client
client_tts = Client("mrfakename/E2-F5-TTS")

def text_to_speech(text, sample):
    result = client_tts.predict(
        ref_audio_input=handle_file(f'input/{sample}.mp3'),
        ref_text_input="",
        gen_text_input=text,
        remove_silence=False,
        cross_fade_duration_slider=0.15,
        speed_slider=1,
        api_name="/basic_tts"
    )
    audio_file = open(result[0], "rb")
    audio_bytes = audio_file.read()
    return audio_bytes

def conversation_predict(input_text):
    """Generate a response for single-turn input using the model."""
    # Tokenize the input text
    input_ids = tokenizer(input_text, return_tensors="pt").input_ids

    # Generate a response with the model
    outputs = model.generate(input_ids, max_new_tokens=2048)

    # Decode and return the generated response
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)

    # Convert the text response to speech using E2-F5-TTS
    audio_bytes = text_to_speech(response, sample="input")
    
    return response, audio_bytes

def respond(
    message: str,
    history: list[tuple[str, str]],
    system_message: str,
    max_tokens: int,
    temperature: float,
    top_p: float,
):
    """Generate a response for a multi-turn chat conversation."""
    # Prepare the messages in the correct format for the API
    messages = [{"role": "system", "content": system_message}]

    for user_input, assistant_reply in history:
        if user_input:
            messages.append({"role": "user", "content": user_input})
        if assistant_reply:
            messages.append({"role": "assistant", "content": assistant_reply})

    messages.append({"role": "user", "content": message})

    response = ""

    # Stream response tokens from the chat completion API
    for message_chunk in client.chat_completion(
        messages=messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = message_chunk["choices"][0]["delta"].get("content", "")
        response += token
        yield response

# Create a Gradio ChatInterface demo
demo = gr.ChatInterface(
    fn=respond,
    additional_inputs=[
        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
)

if __name__ == "__main__":
    demo.launch()