app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import time

# Model configuration
MODEL_NAME = "Qwen/Qwen2-14B-Instruct"

# Initialize model and tokenizer
print("Loading model and tokenizer...")
model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
print("Model and tokenizer loaded!")

def simulate_typing(text, min_chars_per_sec=15, max_chars_per_sec=40):
    """Simulate typing animation with variable speed."""
    full_text = ""
    words = text.split()
    for i, word in enumerate(words):
        full_text += word
        if i < len(words) - 1:
            full_text += " "
        delay = 1 / (min_chars_per_sec + (max_chars_per_sec - min_chars_per_sec) * torch.rand(1).item())
        time.sleep(delay)
        yield full_text

def generate_response(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens=512,
    temperature=0.7,
    top_p=0.95
):
    # Prepare conversation history
    messages = [{"role": "system", "content": system_message}]
    for user_msg, assistant_msg in history:
        if user_msg:
            messages.append({"role": "user", "content": user_msg})
        if assistant_msg:
            messages.append({"role": "assistant", "content": assistant_msg})
    
    messages.append({"role": "user", "content": message})

    # Convert messages to model input format
    text = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True
    )

    # Generate response
    with torch.inference_mode():
        model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
        generated_ids = model.generate(
            **model_inputs,
            max_new_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id
        )
        generated_ids = generated_ids[0, len(model_inputs.input_ids[0]):]
        response = tokenizer.decode(generated_ids, skip_special_tokens=True)

    # Return response with typing animation
    for partial_response in simulate_typing(response):
        yield partial_response

# Custom CSS with typing cursor animation
custom_css = """
@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;600&display=swap');
body, .gradio-container {
    font-family: 'Inter', sans-serif;
}
.typing-cursor::after {
    content: '|';
    animation: blink 1s step-start infinite;
}
@keyframes blink {
    50% { opacity: 0; }
}
"""

# System message
system_message = """You are Qwen 2.5 14B, an advanced AI assistant created by Alibaba Cloud. 
You are knowledgeable, helpful, and strive to provide accurate and comprehensive responses."""

# Gradio chat interface
demo = gr.ChatInterface(
    generate_response,
    additional_inputs=[
        gr.Textbox(
            value=system_message,
            visible=False,
        ),
        gr.Slider(
            minimum=1,
            maximum=2048,
            value=512,
            step=1,
            label="Max new tokens"
        ),
        gr.Slider(
            minimum=0.1,
            maximum=2.0,
            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)"
        ),
    ],
    css=custom_css,
    title="Qwen 2.5 14B Chat",
    description="An advanced AI assistant powered by Qwen 2.5 14B"
)

# Launch the demo
if __name__ == "__main__":
    demo.queue(max_size=40)
    demo.launch(max_threads=40)