app.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

from huggingface_hub import InferenceClient
import gradio as gr

client = InferenceClient('meta-llama/Meta-Llama-3-8B-Instruct')


def funcion(prompt, historial):
    chat = f'<|begin_of_text|><|start_header_id|>system<|end_header_id|>Responde en español con párrafos cortos. Para una mejora de la lectura y comprensión utiliza un formato Markdown enriquecido con # encabezados, ## sub encabezados, **texto enfatizado**, listas, tablas, etc.<|eot_id|>\n\n'
    for usuario, chatbot in historial:
        chat += f'<|start_header_id|>user<|end_header_id|>{usuario}<|eot_id|>\n\n{chatbot}'
    chat += f'<|start_header_id|>user<|end_header_id|>{prompt}<|eot_id|>\n\n'
    kwargs = dict(
        temperature=0.80,
        max_new_tokens=2048,
        top_p=0.95,
        repetition_penalty=1.0,
        seed=1337
    )
    output = ''
    while output == '':
        try:
            stream = client.text_generation(chat, **kwargs, stream=True, details=True, return_full_text=False)
            for response in stream:
                output += response.token.text.replace('<|eot_id|>', '')
                yield output
        except:
            pass
    return output


interfaz = gr.ChatInterface(
        fn=funcion,
        chatbot=gr.Chatbot(
            avatar_images=None,
            container=False,
            show_copy_button=True,
            layout='bubble',
            render_markdown=True,
            line_breaks=True
        ),
        css='h1 {font-size:22px;h2 {font-size:20px;h3 {font-size:18px;h4 {font-size:16px;}',
        autofocus=True,
        fill_height=True,
        analytics_enabled=False,
        submit_btn='Chat',
        stop_btn=None,
        retry_btn=None,
        undo_btn=None,
        clear_btn=None
    )


interfaz.launch()