app.py

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

def get_device():
    """Devuelve 'cuda' si CUDA está disponible (GPU disponible), de lo contrario devuelve 'cpu'."""
    return "cuda" if torch.cuda.is_available() else "cpu"


model_name = "microsoft/Phi-3-mini-128k-instruct"

tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16, trust_remote_code=True)

# Obtener el dispositivo adecuado según la disponibilidad de GPU
device = get_device()
model.to(device)

# Aplicar cuantificación dinámica solo si el dispositivo es CPU
if device == 'cpu':
    model = torch.quantization.quantize_dynamic(
        model, {torch.nn.Linear}, dtype=torch.qint8
    )

pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    device=device,
    trust_remote_code=True
)

@spaces.GPU
def chat_function(message, history, system_prompt, max_new_tokens, temperature):
    prompt = system_prompt
    for msg in history:
        prompt += f'{msg["role"]}: {msg["content"]}\n'
    prompt += f'user: {message}\n'
    
    temp = temperature + 0.1
    outputs = pipeline(
        prompt,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        temperature=temp,
        top_p=0.9,
    )
    generated_text = outputs[0]['generated_text']
    new_text = generated_text[len(prompt):]
    return new_text

chat_interface = gr.ChatInterface(
    fn=chat_function,
    chatbot=gr.Chatbot(height=400),
    textbox=gr.Textbox(placeholder="Enter message here", container=False, scale=7),
    title="Chat with AI Model",
    description="""
The Phi-3-Mini-128K-Instruct is a 3.8 billion-parameter, lightweight, state-of-the-art open model trained using the Phi-3 datasets. This dataset includes both synthetic data and filtered publicly available website data, with an emphasis on high-quality and reasoning-dense properties. The model belongs to the Phi-3 family with the Mini version in two variants 4K and 128K which is the context length (in tokens) that it can support.
https://huggingface.co/microsoft/Phi-3-mini-128k-instruct
    """,
    theme="soft",
    additional_inputs=[
        gr.Textbox(value="Hello!", label="System Prompt", placeholder="Enter a system prompt"),
        gr.Slider(minimum=50, maximum=1000, step=50, value=150, label="Max New Tokens"),
        gr.Slider(minimum=0.0, maximum=1.0, step=0.01, value=0.7, label="Temperature")
    ],
    allow_flagging="never"
)

chat_interface.launch()