app.py

import os
from collections.abc import Iterator
from threading import Thread
import gradio as gr
import spaces
import torch
import edge_tts
import asyncio
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer

MAX_MAX_NEW_TOKENS = 2048
DEFAULT_MAX_NEW_TOKENS = 1024
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

model_id = "prithivMLmods/FastThink-0.5B-Tiny"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype=torch.bfloat16,
)
model.eval()


DESCRIPTION = """
# QwQ Edge 💬
"""

css = '''
h1 {
  text-align: center;
  display: block;
}

#duplicate-button {
  margin: auto;
  color: #fff;
  background: #1565c0;
  border-radius: 100vh;
}
'''

# List of voices
voices = [
    "en-US-JennyNeural",  # @tts1
    "en-US-GuyNeural",    # @tts2
    "en-US-AriaNeural",   # @tts3
    "en-US-JaneNeural",   # @tts4
    "en-US-JasonNeural",  # @tts5
    "en-US-NancyNeural",  # @tts6
    "en-US-TonyNeural",   # @tts7
]


async def text_to_speech(text: str, voice: str, output_file="output.mp3"):
    """Convert text to speech using Edge TTS and save as MP3"""
    communicate = edge_tts.Communicate(text, voice)
    await communicate.save(output_file)
    return output_file


@spaces.GPU
def generate(
    message: str,
    chat_history: list[dict],
    max_new_tokens: int = 1024,
    temperature: float = 0.6,
    top_p: float = 0.9,
    top_k: int = 50,
    repetition_penalty: float = 1.2,
):
    """Generates chatbot response and handles TTS requests"""
    is_tts = message.strip().lower().startswith("@tts")
    tts_index = None

    if is_tts:
        # Extract the number after @tts
        tts_part = message.strip().lower().split()[0]  # Get the @ttsX part
        if len(tts_part) > 8:  # Check if it's @ttsX (e.g., @tts1, @tts2, etc.)
            try:
                tts_index = int(tts_part[8:]) - 1  # Convert to 0-based index
                if tts_index < 0 or tts_index >= len(voices):
                    gr.Warning(f"Invalid TTS voice index. Using default voice.")
                    tts_index = 0
            except ValueError:
                gr.Warning(f"Invalid TTS voice index. Using default voice.")
                tts_index = 0
        else:
            tts_index = 0  # Default to the first voice if no number is provided

        message = message.replace(tts_part, "").strip()  # Remove @ttsX from the message

    conversation = [*chat_history, {"role": "user", "content": message}]

    input_ids = tokenizer.apply_chat_template(conversation, add_generation_prompt=True, return_tensors="pt")
    if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
        input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
        gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
    input_ids = input_ids.to(model.device)

    streamer = TextIteratorStreamer(tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True)
    generate_kwargs = dict(
        {"input_ids": input_ids},
        streamer=streamer,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        top_p=top_p,
        top_k=top_k,
        temperature=temperature,
        num_beams=1,
        repetition_penalty=repetition_penalty,
    )
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    outputs = []
    for text in streamer:
        outputs.append(text)
        yield "".join(outputs)

    final_response = "".join(outputs)

    if is_tts:
        voice = voices[tts_index]  # Select the voice based on the index
        output_file = asyncio.run(text_to_speech(final_response, voice))
        yield gr.Audio(output_file, autoplay=True)  # Return playable audio
    else:
        yield final_response  # Return text response


demo = gr.ChatInterface(
    fn=generate,
    additional_inputs=[
        gr.Slider(label="Max new tokens", minimum=1, maximum=MAX_MAX_NEW_TOKENS, step=1, value=DEFAULT_MAX_NEW_TOKENS),
        gr.Slider(label="Temperature", minimum=0.1, maximum=4.0, step=0.1, value=0.6),
        gr.Slider(label="Top-p (nucleus sampling)", minimum=0.05, maximum=1.0, step=0.05, value=0.9),
        gr.Slider(label="Top-k", minimum=1, maximum=1000, step=1, value=50),
        gr.Slider(label="Repetition penalty", minimum=1.0, maximum=2.0, step=0.05, value=1.2),
    ],
    stop_btn=None,
    examples=[
        ["@tts7 Who is Nikola Tesla, and why did he die?"],
        ["A train travels 60 kilometers per hour. If it travels for 5 hours, how far will it travel in total?"],
        ["Write a Python function to check if a number is prime."],
        ["@tts6 What causes rainbows to form?"],
        ["Rewrite the following sentence in passive voice: 'The dog chased the cat.'"],
        ["@tts4 What is the capital of France?"],
    ],
    cache_examples=False,
    type="messages",
    description=DESCRIPTION,
    css=css,
    fill_height=True,
)

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